Deep Learning-Based Quantitative Measurement Study of Spino-Pelvic Parameters in Adolescent Idiopathic Scoliosis Patients

Deep Learning-Based Quantitative Measurement Study of Spino-Pelvic Parameters in Adolescent Idiopathic Scoliosis Patients | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Research Article Deep Learning-Based Quantitative Measurement Study of Spino-Pelvic Parameters in Adolescent Idiopathic Scoliosis Patients Zhizhen Chen, Dalin Zhu, Jifang Qian, Wenqi Wang, Xiaofei Chen, and 4 more This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-6864546/v1 This work is licensed under a CC BY 4.0 License Status: Under Review Version 1 posted 7 You are reading this latest preprint version Abstract Objective To develop a deep learning (DL) model for the automated measurement of spino-pelvic parameters on whole-spine digital radiography (DR) images of adolescent idiopathic scoliosis (AIS) patients and to evaluate its performance and generalizability. Methods A total of 1348 whole-spine frontal/lateral DR images of AIS patients were collected to train the DL model for predicting fourteen spino-pelvic parameters. The reliability of manual annotation and the model's performance in detecting key points were assessed using Percentage of Correct Keypoints (PCK) values. Differences between the model-predicted parameter values and manual measurements, used as the reference standard, were analyzed using paired t-tests. Further performance evaluations included the mean absolute error (MAE), Pearson correlation coefficient (r), intra-class correlation coefficient (ICC), Bland-Altman plots. Results The DL model demonstrated the ability to automatically detect vertebral bodies and key points. The PCK for detecting vertebral body key points in whole-spine DR images ranged from 82.2–95.3% within a 3-mm threshold. Additionally, the PCK for the left and right femoral heads in whole-spine lateral DR images was 77.9% and 62.3%, respectively. The spino-pelvic parameter values measured by the DL model exhibited a high correlation and agreement with the reference standard (ICC: 0.9–1.0, r: 0.8–1.0, MAE: 1.0–3.7).The Whole frontal model used VF-Net outperformed other networks (one stage HRNet、one stage SCNet、two stage HRNet-HRNet、two stage SCNet-SCNet) in predicting landmarks within a distance threshold of 2.5 to 5 mm;The Whole lateral model used two stage HRNet-HRNet outperformed other networks (VF-Net,one stage HRNet、one stage SCNet、two stage SCNet-SCNet) in predicting landmarks within a distance threshold of 1 to 5 mm. Conclusions The DL model developed in this study can automatically measure spino-pelvic parameters with performance comparable to that of radiologists. This model is expected to provide an automated measurement tool for clinical practice, thereby improving efficiency in diagnosing, monitoring, and treating AIS patients. adolescent idiopathic scoliosis deep learning whole-spine DR image spino-pelvic parameters Figures Figure 1 Figure 2 Figure 3 Figure 4 Figure 5 Figure 6 Figure 7 Figure 8 INTRODUCTION Scoliosis is a complex three-dimensional spinal deformity often characterized by changes in the sagittal and coronal spinal curves, as well as vertebral axis rotation. Adolescent idiopathic scoliosis (AIS) is the most common form of scoliosis, accounting for 75–80% of cases[ 1 ].It is also the most prevalent spinal deformity among adolescents, with a reported prevalence of 1.5–3% in individuals aged 10–18 years, occurring more frequently in females[ 2 ].In China, AIS has emerged as the third most prevalent health concern among adolescents, following obesity and myopia[ 3 ]. Due to the age of onset during adolescence, a period of rapid skeletal development, AIS can progress quickly, leading to clinical manifestations such as rib hump, uneven shoulders, leg-length discrepancies, and cardiopulmonary dysfunction. These symptoms severely impact physical appearance, mental health, and quality of life. Early detection and screening of AIS are critical for preventing disease progression and initiating timely treatment strategies[ 4 ]. Currently, commonly employed methods for scoliosis screening include visual inspection, the Adams forward flexion test, and trunk rotation angle measurements. Patients who screen positive are referred to orthopedic clinics for whole-spine DR imaging to confirm the diagnosis. DR whole-spine imaging is cost-effective, convenient, and efficient. It not only delineates the morphological and structural characteristics of scoliosis but also provides essential information for monitoring disease progression and formulating diagnostic and treatment plans. As such, it is the most widely used imaging technique for assessing scoliosis severity, monitoring its progression, and evaluating spinal balance and treatment outcomes[ 5 ]. The diagnostic criterion for scoliosis is a Cobb angle of ≥ 10° on standing posterior-anterior whole-spine radiographs. The Scoliosis Research Society (SRS) adopted the Cobb angle as the standard method for quantifying scoliosis deformity in 1966, and it remains the most commonly used measurement index[ 6 ]. Spino-pelvic parameters, which describe the three-dimensional alignment of the spine and pelvis, play a crucial role in the clinical diagnosis, evaluation of scoliosis progression, surgical planning, and prognosis of AIS. Manual measurement methods, such as drawing lines on X-ray films with a ruler or protractor or using angle measurement tools in picture archiving and communication systems (PACS), are labor-intensive, require substantial technical experience, and are prone to intra- and inter-observer errors ranging from 2.8° to 8°[ 7 ].Although computer-aided diagnosis (CAD) has improved measurement efficiency through semi-automated methods, manual intervention remains necessary for selecting end vertebrae and determining endplate positions. Variability in endplate selection can result in measurement discrepancies, even among experienced physicians. Moreover, manual operation significantly increases physician workload. With advancements in artificial intelligence (AI), deep learning (DL) has emerged as one of the fastest-growing fields within AI. DL technology has been successfully applied to spine image segmentation, disease diagnosis, and parameter measurement[ 8 , 9 ]. The automated measurement and analysis of spino-pelvic parameters in whole-spine DR images of AIS patients have become a focus of research in DL[ 10 ]. However, existing studies have predominantly focused on the automatic measurement of the Cobb angle in posterior-anterior whole-spine DR images,ignoring whole-spine lateral spino-pelvic parameters[ 11 – 13 ]. Few studies have utilized biplane X-ray images generated by the EOS system in combination with lateral DR images, and the generalizability of these models remains uncertain[ 14 ]. Therefore,this study aimed to develope a DL-based tool for the automated measurement of spino-pelvic parameters in AIS patients using two-dimensional frontal and lateral whole-spine DR images and to evaluate its performance and scalability. The results could provide a quantitative, accurate, and efficient automated measurement technique for clinical diagnosis, treatment, and follow-up of AIS patients. Materials and Methods Datasets The study was conducted in accordance with the Declaration of Helsinki (as revised in 2013). The Ethics Committee of Gansu Provincial Hospital of Traditional Chinese Medicine approved this retrospective study (No. 2020-112-01). The requirement for informed consent was waived because only retrospective imaging data were used. Imaging data of adolescent scoliosis patients who underwent whole-spine digital radiography (DR) at Gansu Provincial Hospital of Traditional Chinese Medicine between January 2018 and October 2022 were retrospectively collected.A total of 1,652 whole-spine frontal and lateral DR radiographs from patients with a clinical diagnosis of AIS were included.Inclusion Criteria:(1)Age: 10–18 years. (2) Whole-spine frontal/lateral radiographs demonstrating the full length of the spine, from the external auditory canal to the femoral head, in patients with a clinical diagnosis of AIS.Exclusion Criteria: (1) Images with significant artifacts or unaligned splices. (2) Incomplete DR images due to positioning issues or deviations from imaging standards. (3) History of spinal surgery or trauma.(4)Presence of metabolic bone disease,spinal tuberculosis,or spinal tumors.(5)Clinical diagnosis of congenital or neuromuscular spinal deformities. Finally,1,348 whole-spine frontal/lateral DR images were ultimately included, comprising 674 frontal and 674 lateral images. They were randomly divided into training, validation, and test sets based on scoliosis severity in proportions of 65%, 15%, and 20%, respectively. The training and test datasets were completely independent to ensure an objective assessment of the model. Annotation of Key Points in Whole-Spine DR Images and Measurement of Spino-Pelvic Parameters To calculate clinically relevant parameters, the naming of each landmark was standardized. The naming rules for key points are illustrated using the L1 vertebral body as an example. For the whole-spine frontal DR images, the key points are named as follows: L1HR refers to the right margin point of the superior edge of the vertebral body; L1FR refers to the right margin point of the inferior edge of the vertebral body; L1HL refers to the left margin point of the superior edge of the vertebral body; and L1FL refers to the left margin point of the inferior edge of the vertebral body. For the whole-spine lateral DR images, the key points are named as follows: L1HA refers to thefrontal margin point of the superior edge of the vertebral body; L1HP refers to the posterior margin point of the superior edge of the vertebral body; L1FA refers to thefrontal margin point of the inferior edge of the vertebral body; and L1FP refers to the posterior margin point of the inferior edge of the vertebral body. The center of the left femoral head is designated as L-FHCE, while the center of the right femoral head is designated as R-FHCE. Figures 1 and 2 depict all key landmark annotations, their specific names, and the parameter measurements[ 15 , 16 ]. Two diagnostic radiologists (R1 and R2), with 10 and 5 years of experience in diagnostic bone and muscle imaging, respectively, were trained in a standardized manner to independently perform pixel-level point labeling of key landmarks in 1,348 whole-spine DR images. This process was conducted using JPHV-specific software designed for model training and validation. To assess intra-observer reliability, the radiologist R2 re-labeled the test set after an interval of six weeks. A third diagnostic radiologist (R3), with 20 years of experience in diagnostic bone and muscle imaging, reviewed all annotations and measurements performed by R1 and R2. R3 corrected any inaccurate annotations and measurements, ensuring the final annotated points and measurements were accurate. These finalized annotations and measurements served as the reference standard for evaluating the performance of the DL in detecting landmarks and measuring parameters. Model Construction DL Modeling of Whole-Spine Frontal and Lateral DR Images in AIS Patients In recent years, through the use of computer vision technology, keypoint localization can help medical experts to analyze image features more accurately and assist in the diagnosis of diseases.Self-calibrated Convolutions Net(SCNet)[ 17 ]and High-Resolution Net ( HRNet) [ 18 ]are the most recent networks applied to keypoint localization. For the deep learning model of whole spine frontal DR images of AIS patients, however, when SCNet and HRNet were used for single-stage keypoint localization in this study, the portion of the spine in the whole spine image was relatively small, and the effective image area was too small, resulting in low localization accuracy. When using their two-stage methods separately, due to the left-right symmetry of the orthotopic vertebrae, the problem of the opposite prediction of the key point positions of a small number of vertebrae occurs. Therefore, in order to solve the above problems, the Vertebra-Focused Landmark Detection Network (VF-Net)[ 19 ]was used in our whole-spine frontal model. For the deep learning model for whole spine lateral DR images of AIS patients, the same problem of very low localization accuracy exists in this study when using SCNet and HRNet for single-stage keypoint localization. Due to the occlusion of the spine by the shoulder joint, arm, and lungs in the whole spine lateral image, the use of VF-Net causes the problem of misalignment of vertebral center prediction, resulting in low localization accuracy. By experimental comparison, HRNet two-stage method has higher localization accuracy than SCNet two-stage method, single-stage HRNet, and VF-Net. Therefore, in our model, the whole spine lateral DR image model uses two-stage HRNet for key point localization. (1) Whole-Spine Frontal DL Model The frontal whole-spine DL model employs the VF-Net backbone network, which output comprises three components: Heatmap, Center Offset, and Corner Offset, as illustrated in Figure S1 [ 19 ]. During training, image augmentation techniques, including horizontal flipping, random rotation(− 30° to + 30°), random scaling, and resizing to 1024 × 512 pixels, were applied. The augmented image was then input into the network to produce the center coordinates, center point offsets, and corner point offsets for each vertebra.For inference, the center coordinates and center point offsets were used to compute the actual coordinates of the vertebral centers. Subsequently, the corner point offsets were used to calculate the coordinates of the four corner points of each vertebra based on the vertebral center coordinates.Based on the calculation method in parameter measurement, the relevant spino-pelvic parameters were then calculated from the coordinates of the predicted landmarks, thus enabling automatic measurement.The model construction flowchart is presented in Fig. 3 . (2) Whole-Spine Lateral DL Model The modeling of lateral whole-spine images was conducted in two stages using the HRNet network[ 20 ] for both stages. Stage 1: The key points at the four corner points of each vertebra were predicted, while the left and right femoral head centers were not considered. For each vertebra, the four corner points were grouped to generate a GT heatmap, resulting in a total of 19 heatmaps corresponding to 19 vertebrae. The network output comprised 19 channels, each predicting the key points for one vertebra. During training, image augmentation methods, including horizontal flipping, random rotation, and random scaling, were applied, after which the image was resized to 1024 × 512 pixels and input into the network. Stage 2: Two HRNet models were trained. The first HRNet model had 4 output channels for predicting the four corner points of each vertebra. The second HRNet model had 2 output channels for predicting the left and right femoral head centers. The approximate locations of the vertebral corner points were obtained using the first-stage network. Each vertebra was then cropped, and the image was resized to 640 × 640 pixels to train the HRNet model with 4 output channels. Similarly, based on the S1 vertebra's position determined by the first-stage network, the region below the S1 vertebra was segmented, resized to 320 × 640 pixels, and input into the HRNet model with 2 output channels. During inference, the exact positions of the four vertebral corner points and the center points of the left and right femoral heads were obtained through the two-stage prediction process. These key points were then integrated into a single image to generate the final prediction of key points. The predicted key points were used to measure spino-pelvic parameters. The model construction flowchart is illustrated in Fig. 4 . Model Performance Evaluation The inter-observer and intra-observer reliability of the landmark annotations were assessed based on the percentage of annotations within landmark-to-landmark distance thresholds of 1 mm, 2 mm, 3 mm, 4 mm, and 5 mm[ 21 ].Annotations corrected by a senior diagnostic radiologist(R3)were used as the reference standard.The model’s performance was evaluated in two dimensions: the accuracy of landmark localization and the consistency of radiological parameter measurements compared with the reference standard on the test set. Specifically, the Percentage of Correct Key Points (PCK) [ 22 ]metric was employed to evaluate the accuracy of all landmarks predicted by the model. Where is an indicator function evaluating to 1 if the location of landmark predicted by the model is within mm of the reference standard , otherwise evaluating to 0. The spino-pelvic parameters were calculated using the coordinates of the key points, as specified in the parameter measurement part. The intra-class correlation coefficient (ICC), the Pearson correlation coefficient (r), and the mean absolute error(MAE) were calculated between the model prediction and reference standard to evaluate the performance of the model[ 23 ].An ICC larger than 0.75 was considered to achieve high consistency between the reference standard and the model in evaluating the radiological parameters. | r |≥0.7 indicated a high correlation. Moreover,the agreement was assessed between model estimated measurements and reference standard using the mean difference, standard deviation (SD), and 95% limit of agreement graphically represented by the Bland-Altman plots. Networks Comparison for Key Point Detection The model performance in detecting all landmarks was compared to different networks(VF-Net、one stage HRNet、one stage SCNet、two stage HRNet-HRNet、two stage SCNet-SCNet) by using PCK on the test sets. All networks received the same training data and strategies for a fair comparison. Statistical Analysis All statistical analyses were performed using Python (Scipy, Statsmodels, and Pingouin). P <0.05 was considered statistically different. The PCK metric was employed to evaluate the reliability of manual labeling and the accuracy of the model in detecting key points. Differences between the reference standard and the model-predicted spino-pelvic parameter values were analyzed using paired t-tests. Additional statistical analyses, including the mean absolute error (MAE), Pearson correlation coefficient (r), intra-class correlation coefficient (ICC), and Bland-Altman plots, were conducted to assess the performance of the model in parameter measurement. Results General Data Distributions A total of 1,348 whole-spine frontal and lateral DR images from 674 adolescent idiopathic scoliosis patients were included in this study for model training, validation, and testing. The dataset comprised 435 images in the training set, 101 in the validation set, and 138 in the testing set. Among all patients, moderate scoliosis was the most prevalent (47.2%), followed by mild scoliosis (37.2%), while severe scoliosis was the least common (15.6%). The study included 198 male and 476 female patients, resulting in a male-to-female ratio of approximately 1:2.4. The proportions of female patients in the training, validation, and testing sets were approximately 69.0%, 75.3%, and 72.5%, respectively. The age group of 13–15 years was the most frequent in all datasets, with proportions of 45.3%, 44.6%, and 40.6% in the training, validation, and testing sets, respectively. No significant differences were observed in the gender and age compositions across the datasets (Table 1 ). Table 1 General Information Distribution of Whole Spine DR Imaging Dataset Scoliosis severity No.of images (sheets) Male Female Average CA (°) Average age (y) No.of images in different age ranges(sheets) 10–12 13–15 16–18 training set light (10°≤CA<20°) 164 68 96 14.4 13.7 58 63 42 middle (20°≤CA<40°) 199 46 153 29.2 14.4 37 95 67 severe(CA ≥ 40°) 72 21 51 56.7 15.2 11 39 22 validation set light (10°≤CA<20°) 31 11 20 14.4 13.4 13 11 7 middle (20°≤CA<40°) 48 9 39 28.5 14.0 13 24 11 severe(CA ≥ 40°) 22 5 17 48.9 14.3 5 10 7 testing set light (10°≤CA<20°) 56 20 36 14.7 13.9 21 19 16 middle (20°≤CA<40°) 71 14 57 28.0 14.0 19 34 18 severe(CA ≥ 40°) 11 4 7 50.3 15 2 3 6 Note: Unless otherwise specified, the data in the table is the number of images. The units in parentheses are the units of measurement data, and the corresponding columns of angle (degree) and average age (y) are the values of measurement data.CA, Cobb angle. Reliability of Manual Annotation The intra- and inter-observer reliability of the annotations for whole-spinefrontal and lateral DR images, labeled by three radiologists (R1, R2, and R3) with varying levels of experience, are shown in Tables 2 and 3 . For intra-observer comparisons, the PCK within the 3-mm threshold was 97.6% for whole-spinefrontal DR images and 90.5% for whole-spine lateral DR images. For inter-observer comparisons, the PCK within the 3-mm threshold was as follows: Whole-spine frontal DR images: 88.8% (R1 vs. R2), 98.7% (R1 vs. R3), and 94.0% (R2 vs. R3). Whole-spine lateral DR images: 86.3% (R1 vs. R2), 96.6% (R1 vs. R3), and 91.8% (R2 vs. R3). Table 2 PCK (%) of manual landmark annotation in whole spine frontal DR images Threshold(mm) 1 2 3 4 5 intra-observer reliability 72.6 94.5 97.6 98.5 98.9 inter-observer reliability R1 vs R2 39.5 76.3 88.8 93.7 96.2 R1 vs R3 81.6 96.3 98.7 99.5 99.7 R2 vs R3 60.7 87.2 94.0 96.7 98.1 PCK, percentage of correct key points. Table 3 PCK (%) of manual landmark annotation in whole spine lateral DR images Threshold(mm) 1 2 3 4 5 intra-observer reliability 61.7 85.3 90.5 92.8 94.1 inter-observer reliability R1 vs R2 42.0 75.9 86.3 91.1 93.3 R1 vs R3 77.4 93.3 96.6 97.7 98.1 R2 vs R3 63.0 86.1 91.8 92.1 94.2 PCK, percentage of correct key points. Landmark Detection Performance The model’s performance in detecting key points in whole-spinefrontal and lateral DR images is presented in Tables 4 and 5 . Within the 3-mm threshold relative to the reference standard: For the whole-spine frontal model, the PCK for vertebral body key points ranged from 82.4–95.3%; For the whole-spine lateral model, the PCK for vertebral body key points ranged from 82.2–92.4%; The PCK for detecting the center points of the left and right femoral heads was 77.9% and 62.8%, respectively. Table 4 PCK (%) of different vertebral bodies in the whole spine frontal model Threshold(mm) 1 2 3 4 5 UV 53.1 63.5 89.2 89.2 89.2 LV 42.1 60.1 82.4 83.8 83.8 AV 55.9 70.3 87.8 87.8 87.8 C7 69.7 91.0 94.5 95.9 95.9 T1 43.3 87.9 95.3 95.9 95.9 S1 22.1 67.6 93.2 94.6 96.0 Mean 47.7 73.4 90.4 91.2 91.4 PCK, Percentage of Correct Key points.UV,upper vertebrae;LV,lower vertebrace;AV,apical vertebrae;C7,7th cervical vertebrae;T1,1st thoracic vertebra;S1,1st sacral vertebra.Mean,The mean Percentage of Correct Key Points. Table 5 PCK (%) of different vertebral bodies and key points in the whole spine lateral model Threshold(mm) 1 2 3 4 5 C7 70.0 85.0 89.5 91.0 92.8 T5 43.3 67.4 82.2 89.3 93.5 T10 69.3 86.2 90.5 91.6 91.7 T12 57.2 82.1 89.8 91.7 92.8 L1 64.8 86.2 90.9 91.9 93.5 L2 64.1 88.3 92.4 94.3 94.5 S1 53.4 75.2 83.3 87.8 90.5 L-FHCE 20.7 52.4 77.9 86.9 89.0 R-FHCE 15.2 45.5 62.8 73.1 79.3 Mean 56.7 78.7 86.5 89.6 91.4 PCK, Percentage of Correct Key points. C7,7th cervical vertebrae;T5,5th thoracic vertebra;T10,10th thoracic vertebra;T12,12th thoracic vertebra;L1,1st lumbar vertebra;L2,2st lumbar vertebra;S1,1st sacral vertebra;L-FHCE,Left femoral head center point;R-FHCE,Right femoral head center point.Mean,The mean Percentage of Correct Key Points. The model demonstrated high accuracy in detecting vertebral and key points in whole-spine frontal and lateral images (Fig. 5). The detected key points in the test set closely approximated the reference standard (Figure S2). Comparison of Networks On the test set, the PCK performance of the proposed models was compared to other DL models. For the whole-spine frontal model, the PCK within the 2.5–5 mm threshold was higher for VF-Net (proposed model) than for the one-stage HRNet, one-stage ScNet, two-stage HRNet, and two-stage ScNet. The highest PCKs for each model were as follows: 93.5% (VF-Net), 67.9% (one-stage HRNet), 66.4% (one-stage ScNet), 78.3% (two-stage HRNet), and 79.3% (two-stage ScNet). For the whole-spine lateral model, the PCK within the 1–5 mm threshold was highest for the two-stage HRNet (proposed model), followed by the two-stage ScNet, one-stage ScNet, one-stage HRNet, and VF-Net. The highest PCKs for each model were as follows: 88.1% (two-stage HRNet), 85.3% (two-stage ScNet), 83.7% (one-stage ScNet), 86.3% (one-stage HRNet), and 85.9% (VF-Net) (Fig. 6 ). Model Measurement Performance No significant differences were observed between the predicted spino-pelvic parameter values of the whole-spine frontal/lateral models and the reference standard (p > 0.05) (Tables 6 and 7 ). Overall, the model’s predicted spino-pelvic parameter values were consistent and reliable when compared to the reference standard(Whole-spine frontal model: ICC: 0.9–1.0, r: 0.8–1.0, MAE: 1.0–3.7;Whole-spine lateral model: ICC: 0.9–1.0, r: 0.9–1.0, MAE: 1.0–3.5). Bland-Altman plots and regression analyses further illustrated the differences and correlations between the model predictions and the reference standard (Figs. 7 and 8 ). Table 6 Comparison between the model and the reference standard for the measurement of spino-pelvic parameter(whole spinefrontal) Parameters reference standard predicted value P ICC[95%CI] r MAE CVA(mm) 12.9 ± 9.0 12.6 ± 9.1 0.9 1.0[0.9,1.0] 1.0 1.0 AVT(mm) 20.6 ± 15.0 20.6 ± 15.1 1.0 1.0[0.9,1.0] 1.0 1.1 Cobb angle(°) Light (10°≤CA<20°) 14.6 ± 2.8 15.1 ± 3.8 0.6 0.9[0.8,0.9] 0.9 1.0 Middle (20°≤CA<40°) 27.3 ± 5.3 28.0 ± 6.0 0.4 0.9[0.8,0.9] 0.9 2.4 Severe (CA ≥ 40°) 50.3 ± 10.5 52.9 ± 14.8 0.7 0.9[0.6,0.9] 0.8 3.7 Total_CA(°) 22.2 ± 9.9 22.8 ± 10.2 0.6 0.9[0.9,1.0] 0.9 1.8 T1 TA(°) 4.0 ± 4.2 3.9 ± 4.7 0.8 0.9[0.9,1.0] 0.9 1.3 Data are expressed as the mean ± SD.P < 0.05 was considered statistically different.ICC (95% CI), intra-class correlation coefficient (95% confidence interval); r, Pearson correlation coefficient; MAE, the mean absolute error.CVA,coronal vertical axis;AVT,apical vertebral translation;CA,Cobb angle;T1 TA,T1 tital Angle. Table 7 Comparison between the model and the reference standard for the measurement of spino-pelvic parameter(whole spine lateralization) Parameters reference standard predicted value P ICC[95%CI] r MAE TK(°) 23.5 ± 9.8 23.3 ± 10.1 0.9 0.9[0.8,0.9] 0.9 3.5 TLK(°) 8.4 ± 6.1 8.5 ± 6.3 0.9 0.9[0.8,0.9] 0.9 2.2 LL(°) 47.1 ± 10.7 47.8 ± 11.3 0.6 1.0[0.9,1.0] 1.0 2.5 SVA(°) 21.8 ± 16.9 21.9 ± 17.0 1.0 1.0[1.0,1.0] 1.0 1.1 SS(°) 40.6 ± 21.0 40.4 ± 21.2 1.0 1.0[1.0,1.0] 1.0 1.5 PI(°) 47.1 ± 10.2 46.3 ± 10.8 0.6 1.0[1.0,1.0] 1.0 1.9 PT(°) 10.7 ± 6.9 10.1 ± 6.9 0.5 1.0[1.0,1.0] 1.0 1.0 Data are expressed as the mean ± SD.P < 0.05 was considered statistically different.ICC (95% CI), intra-class correlation coefficient (95% confidence interval); r, Pearson correlation coefficient; MAE, the mean absolute error.TK,thoracic kyphosis;TLK,thoracolumbar kyphosis;LL,lumbar lordosis;SVA,sagital vetebare axis;SS,sacral slope;PI,pelvic incidence;PT,pelvic tilt. Discussion Quantitative, accurate, and efficient measurement of spino-pelvic parameters in adolescent idiopathic scoliosis (AIS) patients is essential for disease diagnosis, treatment planning, and prognosis assessment[24, 2 5]. In this study, we developed a deep learning (DL) model for the automatic measurement of spino-pelvic parameters in AIS patients. The model can accurately locate vertebrae and key points on whole-spine frontal and lateral DR images and automatically measure spino-pelvic parameters, achieving predictions comparable to those of senior diagnostic radiology reviewers. Traditionally, manual measurement of spino-pelvic parameters relies on the subjective judgment and clinical experience of observers, making intra- and inter-observer variations inevitable[ 26 , 27 ]. Errors in the measurement of spino-pelvic parameters have been studied. It has been reported in the literature that the variation in Cobb angle measurements by different observers ranged from 2° to 11°[ 28 ].Some scholars also used computer-assisted techniques to measure spondylopelvic parameters in whole spine DR images to study and analyze the differences in manual measurements. The study showed that although computer-assisted semi-automated measurement tools are more advantageous than manual measurement in measuring spino-pelvic parameters[ 27 ], the measurement process requires manual selection of vertebrae and marking of anatomical positions for the measurement of relevant parameters, and manual participation is unavoidable. Even when the same diagnostic criteria are followed, experienced diagnostic radiologists may arrive at different assessment results.Recent advances in DL techniques have provided new approaches capable of outperforming professional human operators in tasks such as image classification, target detection, and landmark localization. Several studies have applied DL to scoliosis screening, the prediction of disease progression and brace treatment outcomes, Lenke classification, and the selection of fused vertebral segments[ 29 , 30 ]. However, these studies primarily focused on the automatic measurement of Cobb angle using DL techniques, overlooking other critical spinal frontal planar, sagittal planar, and pelvic parameters. In contrast, the DL model developed in this study based on whole-spine frontal and lateral DR images demonstrated the following: (1) The model’s performance in detecting vertebrae key points was comparable to the results of diagnostic radiology reviewers. Within a 3-mm threshold of the reference standard, the model’s PCK for vertebrae key points on frontal and lateral DR images ranged from 82.2–95.3%, while the PCK for predicting the center points of the left and right femoral heads was 77.9% and 62.8%, respectively(Tables 4 and 5 ;Figure 5). (2) The spino-pelvic parameters measured by the DL model on frontal and lateral DR images exhibited high correlation and consistency with the reference standards (Tables 6 and 7 ; Figs. 7 and 8 ). (3) For whole-spine frontal DR images, the VF-Net used in this study demonstrated superior accuracy in detecting vertebrae key points within the 2.5–5 mm threshold compared to SCNet and HRNet. For whole-spine lateral DR images, the two-stage HRNet method proposed in this study outperformed SCNet, VF-Net, and single-stage HRNet in detecting vertebrae and key points within the 1–5 mm threshold (Fig. 6 ). To ensure accurate parameter measurements, key points were identified based on corrections provided by senior diagnostic radiology reviewers, which served as the reference standard. Chen et al.[ 31 ]reported that data with a mean inter-observer landmark distance within 3 mm is suitable for clinical analysis. Consistent with this, the results of our study showed that most inter-observer distances fell within the 3-mm threshold, confirming the reliability of the labeling and auditing results from the three radiologists (Tables 2 and 3 ).Furthermore, within the 3-mm threshold, the model-predicted PCK values for the upper and lower end vertebrae and the apical vertebrae were 89.2%, 82.4%, and 87.8%, respectively, for whole-spine frontal DR images, which were comparable to and in some cases slightly higher than the results reported in existing studies[ 32 ]. Additionally, the DL model performed well in detecting the T1, C7, and S1 vertebrae, achieving PCK values of 95.3%, 94.5%, and 93.2%, respectively, within the 3-mm threshold. For lateral DR images, the model’s accuracy in detecting vertebrae and key points was slightly lower than that for frontal DR images, with mean PCK values of 90.4% and 86.5%, respectively. This difference can be attributed to challenges such as the overlap of lung tissue and thoracic bony structures, air-filled abdominal intestinal tubes, and pelvic anatomical structures in lateral images, which hinder precise identification of vertebrae key points.The model achieved the lowest PCK values for the femoral head centers, followed by the T5 vertebra (PCK: 82.2%). This finding aligns with previous studies[ 33 ]and is primarily due to the lower image contrast in the pelvic region of lateral DR images, which impacts the detection of femoral head centers. Additionally, anatomical interference from structures such as the scapula, clavicle, costovertebral joints, aorta, and mediastinum contributes to boundary blurring of the T5 vertebra, posing challenges even for diagnostic radiologists. These limitations are objective, unavoidable, and independent of the DL model itself. Several studies have explored DL techniques for automated spinal curvature measurements in AIS patients, achieving promising results. For example, Horng et al.[ 34 ] validated their model’s performance within a Cobb angle range of 0–20.1° but did not evaluate it for Cobb angles exceeding 20.1°. Similarly, Zhang et al.[ 35 ]developed a vertebral angle-point detection method for Cobb angle measurement but did not employ statistical methods to assess its consistency with manual measurements. Wu et al. [ 36 ] proposed a Multi-View Correlation Network (MVC-Net) for fully automated spinal curvature assessment, achieving a mean absolute error (MAE) of 4.0° for the Cobb angle. Sardjono et al.[ 37 ]tested a DL model for automated Cobb angle measurement, reporting an MAE of 3.9°. The clinically acceptable error range for automated Cobb angle measurements is 3–5° [ 38 ]. Compared to these studies, the DL model developed in this study demonstrated superior performance, particularly in predicting severe Cobb angles. The MAEs for mild, moderate, severe, and overall Cobb angles were 1.0°, 2.4°, 3.7°, and 1.8°, respectively, which are lower than previously reported values. Huang et al.[ 39 ]applied a convolutional neural network (CNN) to 300 whole-spine frontal DR images of AIS patients, reporting intra-class correlation coefficients (ICCs) of 0.6, 0.9, and 0.8 for mild, moderate, and severe Cobb angles, respectively. In contrast, the ICCs of mild, moderate, severe, and overall Cobb angles for the DL model in this study were same (ICC = 0.9), indicating higher reliability and accuracy.Moreover, the model demonstrated excellent performance in predicting coronal vertical axis (CVA) and apical vertebral translation (AVT) parameters, with MAEs of 1.0 mm and 1.1mm, respectively, and ICCs of 1.0 for both parameters, comparable to the results reported by Wu et al. [ 32 ]for multi-stage integrated network models. Additionally, T1 tilt angle measurements were included in this study, showing high accuracy with an MAE of 1.3°, an ICC of 0.9, and a Pearson correlation coefficient (r) of 0.9, approaching the results obtained by senior diagnostic radiology reviewers. Combining coronal and sagittal measurements of spino-pelvic parameters to provide a comprehensive analysis at the AIS patient level is a notable advantage. Zerouali et al. [ 40 ]developed a DL model for the automatic assessment of spino-pelvic coronal and sagittal parameters based on data from 100 pediatric and adult scoliosis patients. The model demonstrated good consistency and accuracy (ICC ≥ 0.9; MAE ≤ 4.4° or 2.7 mm) for most parameters, except for thoracic kyphosis (TK), where the ICC was 0.58 and the MAE was 8.7°. Wu et al.[ 36 ]proposed the MVC-Net network architecture, which overcomes the challenges of sternal overlap in whole-spine lateral images. Their model achieved a mean absolute error (MAE) of 4.1° in predicting the lateral Cobb angle. Similarly, although the performance of our DL model in predicting TK for whole-spine lateral images was slightly inferior to that for other parameters, consistent with existing studies, the MAE for TK was 3.5°, with ICC and Pearson correlation coefficient (r) values of 0.9. This indicates that the error is smaller, and the consistency and correlation are higher than those reported for the aforementioned DL models. In addition, Galbusera et al.[ 14 ]developed a DL model using biplanar radiographs acquired via the EOS imaging system, with a standard deviation range of 2.7° (Pelvic Tilt, PT) to 11.5° (L1-L5 spinalfrontal convexity angle) for predicting sagittal spino-pelvic parameters. By comparison, the DL model developed in the present study achieved a standard deviation range of 1.7° (Pelvic Tilt,PT) to 4.6° (thoracic kyphosis,TK), highlighting its superior performance in measuring sagittal spino-pelvic parameters in AIS patients. With the advancement and optimization of DL network models, SCNet and HRNet, which are state-of-the-art methods for detecting key points in human posture estimation tasks, exhibit low accuracy when locating vertebrae in whole-spine frontal DR images using single-stage approaches. Furthermore, incorrect key point detection occurs when their two-stage methods are applied separately. To address these limitations, the present study adopted the VF-Net single-stage approach for the whole-spine frontal DL model, incorporating a ResNet34 network as the encoder. This strategy improves the detection accuracy of vertebrae and key points by employing vertebrae center heatmaps, center offsets,and corner offsets. Fully automated detection of vertebral key points demonstrated greater accuracy, particularly within the range of 2.5-5 mm. Based on the accurate identification of vertebrae key points, any two vertebrae from the 17 included (12 thoracic and 5 lumbar) were selected to calculate the angle between the upper endplate of the superior vertebra and the lower endplate of the inferior vertebra. The largest angle obtained was identified as the Cobb angle, and the corresponding vertebrae were labeled as the upper and lower end vertebrae[ 41 ]. Furthermore, the distance between the center of each vertebra and the center sacral vertical line within the curvature range was calculated, with the largest distance defined as the apical vertebral translation (AVT) and the corresponding vertebra identified as the apical vertebra. Additionally, the model automatically measured other spino-pelvic parameters of the whole spine in the frontal plane. Inter-model comparisons revealed that the two-stage HRNet model demonstrated superior performance in detecting key points in whole-spine lateral images compared to other models. In this study, the two-stage HRNet method, combined with end-to-end learning, effectively maintained high resolution, enabling the automatic localization of vertebral body corner points and bilateral femoral head centers in whole-spine lateral images.The model exhibited a significant advantage over SCNet, VF-Net, and single-stage HRNet in detecting vertebral bodies and key points within a threshold range of 1–5 mm. This facilitated the automatic measurement of spino-pelvic parameters in the lateral plane. Although the performance of the whole-spine lateral model for detecting bilateral femoral head centers at the 3-mm threshold was lower than that for vertebrae detection, this discrepancy had minimal impact on pelvic parameter measurements. Specifically, the distance between the midpoints of the bilateral femoral head center lines and the midpoint of the first sacral upper endplate was approximately 15 cm, reducing the influence of center point detection errors on pelvic parameter assessments. There are several limitations to this study. First, this is a single-center retrospective study, which lacks an external validation dataset. Variations in image quality may limit the performance of the DL model. Future work will focus on developing DL models using multicenter databases to address this limitation. Second, the current DL model may not accurately predict spino-pelvic parameters in patients with vertebral sequence abnormalities, such as lumbosacral transitional vertebrae or congenital vertebral anomalies. The accuracy of our DL method in predicting landmarks relies heavily on the correct identification of vertebrae and their boundary corner points. To address this issue, it may be necessary to design an algorithm that determines the number of vertebrae before landmark localization, while simultaneously increasing the diversity of scoliosis cases in the training dataset. Third, the current study includes a predominance of AIS patients with moderate scoliosis, while the number of patients with severe scoliosis remains relatively small. Consequently, the training dataset may not fully represent complex clinical settings. Future studies will aim to incorporate a broader range of cases to improve model generalizability. In conclusion, this study developed and validated an automated DL-based measurement model for spino-pelvic parameters on whole-spine frontal and lateral DR images in AIS patients. The model demonstrated performance comparable to senior diagnostic radiology reviewers in detecting vertebrae, identifying key points, and measuring most spino-pelvic parameters. Furthermore, its performance was close to or exceeded that of previously reported DL models. In future research, we plan to conduct multicenter studies and collect external validation datasets encompassing a wide range of scoliosis types and severities. This will enhance the robustness of the model and provide a reliable, efficient tool for intelligent assessment in clinical practice, optimizing the diagnostic and therapeutic workflow for AIS patients. Declarations Acknowledgments This work was supported by the General Project of the Joint Research Fund of Gansu Province (23JRRA1542),Gansu Provincial Hospital;National Natural Science Foundation of China (NSFC)(82360358),Gansu Provincial Hospital. Footnote Conflicts Of Interest All authors have completed the ICMJE uniform disclosure form. Guohua Cheng is a consultant of Hangzhou Jianpei Technology Co., Ltd.Linyang He is an employee of Hangzhou Jianpei Technology Co., Ltd.The authors have no conflicts of interest to declare. Ethical Statement The authors are accountable for all aspects of the work in ensuring that questions related to the accuracy or integrity of any part of the work are appropriately investigated and resolved. The study was conducted in accordance with the Declaration of Helsinki (as revised in 2013). The study was approved by Gansu Provincial Hospital of Traditional Chinese Medicine approved this retrospective study (No. 2020-112-01). The requirement for informed consent was waived because retrospective imaging data were used. Human Ethics and Consent to Participate declarations : not applicable Author Contribution Contributions: (I) Conception and design: Z Chen; (II) Administrative support: D Zhu; (III) Provision of study materials or patients: J Qian; (IV) Collection and assembly of data: W Wang; (V) Data analysis and interpretation: G Cheng; (VI) Manuscript writing: All authors; (VII) Final approval of manuscript: All authors. References Kuznia AL, Hernandez AK, Lee LU (2020) Adolescent Idiopathic Scoliosis: Common Questions and Answers. AM FAM PHYSICIAN 101:19-23 Horne JP, Flannery R, Usman S (2014) Adolescent idiopathic scoliosis: diagnosis and management. AM FAM PHYSICIAN 89:193 Lv Z, Lv W, Wang L, Ou J (2023) Development and validation of machine learning-based models for prediction of adolescent idiopathic scoliosis: A retrospective study. MEDICINE 102:e33441. 10.1097/MD.0000000000033441*10.1097/MD.0000000000033441 Altaf F, Gibson A, Dannawi Z, Noordeen H (2013) Adolescent idiopathic scoliosis. BMJ-BRIT MED J 346:f2508. 10.1136/bmj.f2508*10.1136/bmj.f2508 García-Cano E, Arámbula Cosío F, Duong L, Bellefleur C, Roy-Beaudry M, Joncas J, Parent S, Labelle H (2018) Prediction of spinal curve progression in Adolescent Idiopathic Scoliosis using Random Forest regression. COMPUT BIOL MED 103:34-43. 10.1016/j.compbiomed.2018.09.029*10.1016/j.compbiomed.2018.09.029 Thompson GH (2010) Scoliosis Research Society. SPINE 35:134-137 Zhou Z, Zhu J, Yao C (2023) Vertebral Center Points Locating and Cobb Angle Measurement Based on Deep Learning. Applied Sciences 13:3817. 10.3390/app13063817*10.3390/app13063817 Gaonkar B, Villaroman D, Beckett J, Ahn C, Macyszyn L (2019) Quantitative Analysis of Spinal Canal Areas in the Lumbar Spine: An Imaging Informatics and Machine Learning Study. AM J NEURORADIOL 40:1586-1591 Wimmer M, Major D, Novikov AA, Bühler K (2018) Fully automatic cross-modality localization and labeling of vertebral bodies and intervertebral discs in 3D spinal images. International Journal of Computer Assisted Radiology & Surgery:1-13 San Román Gaitero A, Shoykhet A, Spyrou I, Stoorvogel M, Vermeer L, Schlösser TPC (2023) Imaging Methods to Quantify the Chest and Trunk Deformation in Adolescent Idiopathic Scoliosis: A Literature Review. HEALTHCARE-BASEL 11:1489. 10.3390/healthcare11101489*10.3390/healthcare11101489 Kokabu T, Kanai S, Kawakami N, Uno K, Kotani T, Suzuki T, Tachi H, Abe Y, Iwasaki N, Sudo H (2021) An algorithm for using deep learning convolutional neural networks with three dimensional depth sensor imaging in scoliosis detection. Spine J 21:980-987. 10.1016/j.spinee.2021.01.022*10.1016/j.spinee.2021.01.022 Chen B, Xu Q, Wang L, Leung S, Chung J, Li S (2019) An Automated and Accurate Spine Curve Analysis System. IEEE ACCESS 7:124596-124605. 10.1109/ACCESS.2019.2938402*10.1109/ACCESS.2019.2938402 Dubost F, Collery B, Renaudier A, Roc A (2020) Automated Estimation of the SpinalCurvature via Spine Centerline Extraction with Ensembles of Cascaded Neural Networks. Computational Methods and Clinical Applications for Spine Imaging.CSI 2019. Lecture Notes in Computer Science 11963:88-94. 10.1007/978-3-030-39752-4_10*10.1007/978-3-030-39752-4_10 Galbusera F, Niemeyer F, Wilke HJ, Bassani T, Casaroli G, Anania C, Costa F, Brayda-Bruno M, Sconfienza LM (2019) Fully automated radiological analysis of spinal disorders and deformities: a deep learning approach. EUR SPINE J 28:951-960. 10.1007/s00586-019-05944-z*10.1007/s00586-019-05944-z Singh R, Yadav SK, Sood S, Yadav RK, Rohilla R (2018) Spino-pelvic radiological parameters in normal Indian population. SICOT-J 4:14 Yao J, Dong B, Sun J, Liu J, Zhang J (2020) Accuracy and Reliability of Computer゛ided Anatomical Measurements for Vertebral Body and Disc Based on Computed Tomography Scans. ORTHOP SURG Lee CY, Xie S, Gallagher P, Zhang Z, Tu Z (2014) Deeply-Supervised Nets. Eprint Arxiv:562-570. 10.48550/arXiv.1409.5185*10.48550/arXiv.1409.5185 Wang J, Sun K, Cheng T, Jiang B, Deng C, Zhao Y, Liu D, Mu Y, Tan M, Wang X, Liu W, Xiao B (2021) Deep High-Resolution Representation Learning for Visual Recognition. IEEE T PATTERN ANAL 43:3349-3364. 10.1109/TPAMI.2020.2983686*10.1109/TPAMI.2020.2983686 Yi JR, Wu PX, Huang QY, Hui Q (2020) VERTEBRA-FOCUSED LANDMARK DETECTION FOR SCOLIOSIS ASSESSMENT. In. Rutgers University, Piscataway. Sun K, Xiao B, Liu D, Wang J (2019) Deep High-Resolution Representation Learning for Human Pose Estimation. In: 2019 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). Cornell University Library. Payer C, Tern D, Bischof H, Urschler M (2019) Integrating Spatial Configuration into Heatmap Regression Based CNNs for Landmark Localization. MED IMAGE ANAL Li T, Wang Y, Qu Y, Dong R, Kang M, Zhao J (2022) Feasibility study of hallux valgus measurement with a deep convolutional neural network based on landmark detection. SKELETAL RADIOL 51:1235-1247 Yang W, Ye Q, Ming S, Hu X, Jiang Z, Shen Q, He L, Gong X (2020) Feasibility of automatic measurements of hip joints based on pelvic radiography and a deep learning algorithm. EUR J RADIOL 132:109303. 10.1016/j.ejrad.2020.109303*10.1016/j.ejrad.2020.109303 Bernstein P, Metzler J, Weinzierl M, Seifert C, Kisel W, Wacker M (2021) Radiographic scoliosis angle estimation: spline-based measurement reveals superior reliability compared to traditional COBB method. EUR SPINE J 30:676-685. 10.1007/s00586-020-06577-3*10.1007/s00586-020-06577-3 Soffer S, Ben-Cohen A, Shimon O, Amitai MM, Greenspan H, Klang E (2019) Convolutional Neural Networks for Radiologic Images: A Radiologist's Guide. RADIOLOGY 290:590-606. 10.1148/radiol.2018180547*10.1148/radiol.2018180547 Prestigiacomo FG, Hulsbosch MHHM, Bruls VEJ, Nieuwenhuis JJ (2022) Intra- and inter-observer reliability of Cobb angle measurements in patients with adolescent idiopathic scoliosis. SPINE DEFORM 10:79-86. 10.1007/s43390-021-00398-0*10.1007/s43390-021-00398-0 Wu W, Liang J, Du Y, Tan X, Xiang X, Wang W, Ru N, Le J (2014) Reliability and reproducibility analysis of the Cobb angle and assessing sagittal plane by computer-assisted and manual measurement tools. BMC MUSCULOSKEL DIS 15:33. 10.1186/1471-2474-15-33*10.1186/1471-2474-15-33 Cheung J, Wever DJ, Veldhuizen AG, Klein JP, Verdonck B, Nijlunsing R, Cool JC, Van Horn JR (2002) The reliability of quantitative analysis on digital images of the scoliotic spine. European spine journal 11:535-542 Yang J, Zhang K, Fan H, Huang Z, Xiang Y, Yang J, He L, Zhang L, Yang Y, Li R (2019) Development and validation of deep learning algorithms for scoliosis screening using back images. COMMUN BIOL 2:390 Phan P, Mezghani N, Wai EK, Guise JD (2013) Artificial neural networks assessing adolescent idiopathic scoliosis: comparison with Lenke classification - ScienceDirect. Spine J 13:1527-1533. 10.1016/j.spinee.2013.07.449.*10.1016/j.spinee.2013.07.449. Chen H, Lin C, Wu C, Wang C, Sun Y (2010) Automatic Insall–Salvati ratio measurement on lateral knee x-ray images using model-guided landmark localization. PHYS MED BIOL 55:6785-6800. 10.1088/0031-9155/55/22/012*10.1088/0031-9155/55/22/012 Wu C, Meng G, Lian J, Xu J, Gao M, Huang C, Zhang S, Zhang Y, Yu Y, Wang H, He Y, Li X, Li Z (2022) A multi-stage ensemble network system to diagnose adolescent idiopathic scoliosis. EUR RADIOL. 10.1007/s00330-022-08692-9*10.1007/s00330-022-08692-9 Yeh Y, Weng C, Huang Y, Fu C, Tsai T, Yeh C (2021) Deep learning approach for automatic landmark detection and alignment analysis in whole-spine lateral radiographs. SCI REP-UK 11. 10.1038/s41598-021-87141-x*10.1038/s41598-021-87141-x Horng M, Kuok C, Fu M, Lin C, Sun Y (2019) Cobb Angle Measurement of Spine from X-Ray Images Using Convolutional Neural Network. COMPUT MATH METHOD M 2019:1-18. 10.1155/2019/6357171*10.1155/2019/6357171 Zhang K, Xu N, Yang G, Wu J (2019) An Automated Cobb Angle Estimation Method Using Convolutional Neural Network with Area Limitation. In: Medical Image Computing and Computer-Assisted Intervention-MICCAI 2019. Springer International Publishing, Switzerland. pp. 775-783. Wu H, Bailey C, Rasoulinejad P, Li S (2018) Automated comprehensive Adolescent Idiopathic Scoliosis assessment using MVC-Net. MED IMAGE ANAL 48:1-11. https://doi.org/10.1016/j.media.2018.05.005*https://doi.org/10.1016/j.media.2018.05.005 Sardjono TA, Wilkinson MH, Veldhuizen AG, van Ooijen PM, Purnama KE, GJ. V (2013) Automatic Cobb Angle Determination From Radiographic Images. SPINE 38:E1256-E1262 Lechner R, Putzer D, Dammerer D, Liebensteiner M, Bach C, Thaler M (2017) Comparison of two- and three-dimensional measurement of the Cobb angle in scoliosis. INT ORTHOP 41:957-962. 10.1007/s00264-016-3359-0*10.1007/s00264-016-3359-0 Huang X, Luo M, Liu L, Wu D, You X, Deng Z, Xiu P, Yang X, Zhou C, Feng G, Wang L, Zhou Z, Fan J, He M, Gao Z, Pu L, Wu Z, Zhou Z, Song Y, Huang S (2022) The Comparison of Convolutional Neural Networks and the Manual Measurement of Cobb Angle in Adolescent Idiopathic Scoliosis. GLOB SPINE J 0:1-10. 10.1177/21925682221098672*10.1177/21925682221098672 Zerouali M, Parpaleix A, Benbakoura M, Rigault C, Champsaur P, Guenoun D (2023) Automatic deep learning-based assessment of spinopelvic coronal and sagittal alignment. DIAGN INTERV IMAG 15:1-8. https://doi.org/10.1016/j.diii.2023.03.003*https://doi.org/10.1016/j.diii.2023.03.003 Sun Y, Xing YZ, Zhao ZA, Meng X, Xu G, Hai Y (2022) Comparison of manual versus automated measurement of Cobb angle in idiopathic scoliosis based on a deep learning keypoint detection technology. EUR SPINE J 31:1969-1978. 10.1007/s00586-021-07025-6*10.1007/s00586-021-07025-6 Additional Declarations No competing interests reported. Supplementary Files supplementalmaterial.docx Cite Share Download PDF Status: Under Review Version 1 posted Reviews received at journal 30 Jul, 2025 Reviewers agreed at journal 20 Jul, 2025 Reviewers invited by journal 02 Jul, 2025 Editor invited by journal 23 Jun, 2025 Editor assigned by journal 12 Jun, 2025 Submission checks completed at journal 12 Jun, 2025 First submitted to journal 10 Jun, 2025 You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-6864546","acceptedTermsAndConditions":true,"allowDirectSubmit":false,"archivedVersions":[],"articleType":"Research Article","associatedPublications":[],"authors":[{"id":480120028,"identity":"5efc7c2e-1577-4290-9f49-ec786f8a01d9","order_by":0,"name":"Zhizhen Chen","email":"","orcid":"","institution":"Medical Imaging Center of Gansu Provincial Maternity and Child-care Hospital(Gansu Provincial Central Hospital)","correspondingAuthor":false,"prefix":"","firstName":"Zhizhen","middleName":"","lastName":"Chen","suffix":""},{"id":480120029,"identity":"75ab19af-9e24-4b7f-b7f7-e993ca818584","order_by":1,"name":"Dalin Zhu","email":"","orcid":"","institution":"Medical Imaging Center of Gansu Provincial Maternity and Child-care Hospital(Gansu Provincial Central Hospital)","correspondingAuthor":false,"prefix":"","firstName":"Dalin","middleName":"","lastName":"Zhu","suffix":""},{"id":480120030,"identity":"a4ac26f5-7b7a-4b49-abd7-909a5f06a51f","order_by":2,"name":"Jifang Qian","email":"","orcid":"","institution":"Medical Imaging Center of Gansu Provincial Maternity and Child-care Hospital(Gansu Provincial Central Hospital)","correspondingAuthor":false,"prefix":"","firstName":"Jifang","middleName":"","lastName":"Qian","suffix":""},{"id":480120031,"identity":"4aec1120-5166-4836-babb-e32196391ad0","order_by":3,"name":"Wenqi Wang","email":"","orcid":"","institution":"Gansu Provincial Hospital of Traditional Chinese Medicine","correspondingAuthor":false,"prefix":"","firstName":"Wenqi","middleName":"","lastName":"Wang","suffix":""},{"id":480120032,"identity":"8b521df1-ec20-4636-8f0e-131bf28e59c2","order_by":4,"name":"Xiaofei Chen","email":"","orcid":"","institution":"Gansu Provincial Hospital of Traditional Chinese Medicine","correspondingAuthor":false,"prefix":"","firstName":"Xiaofei","middleName":"","lastName":"Chen","suffix":""},{"id":480120034,"identity":"e94408e8-573d-4e61-81f4-42c43ec302ed","order_by":5,"name":"Fuwen Dong","email":"","orcid":"","institution":"Gansu Provincial Hospital of Traditional Chinese Medicine","correspondingAuthor":false,"prefix":"","firstName":"Fuwen","middleName":"","lastName":"Dong","suffix":""},{"id":480120036,"identity":"5c76acb3-7291-45c1-b3dc-97984519203e","order_by":6,"name":"Guohua Cheng","email":"","orcid":"","institution":"Hangzhou Jianpei Technology Co., Ltd","correspondingAuthor":false,"prefix":"","firstName":"Guohua","middleName":"","lastName":"Cheng","suffix":""},{"id":480120040,"identity":"b9cebcd7-1e66-415d-ad25-2c96e1540c10","order_by":7,"name":"Linyang He","email":"","orcid":"","institution":"Hangzhou Jianpei Technology Co., Ltd","correspondingAuthor":false,"prefix":"","firstName":"Linyang","middleName":"","lastName":"He","suffix":""},{"id":480120041,"identity":"a806741b-23cf-44e2-9c13-7931f0c6db8c","order_by":8,"name":"Sheng Zhou","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAAmUlEQVRIiWNgGAWjYBACPhDxgUGCBC1sQMw4g2QtzDykuIuBjf/44882fyzy+BuYHz66QZwtB9Kkc9skiiUOsBkb5xClhbHhGHNug0RiwwEeNmnitDAzNn+2+COROJ94LWzMDNIMbBKJG4jXwsPGJtnbJpG48TCxfuEHhtiHH3/qEucdb374mCgtCMBMmvJRMApGwSgYBfgAADiSKDDh2ck3AAAAAElFTkSuQmCC","orcid":"","institution":"Gansu Provincial Hospital","correspondingAuthor":true,"prefix":"","firstName":"Sheng","middleName":"","lastName":"Zhou","suffix":""}],"badges":[],"createdAt":"2025-06-10 15:23:21","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-6864546/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-6864546/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":86141032,"identity":"b755fe99-911a-4727-8d06-26d5ff378a2a","added_by":"auto","created_at":"2025-07-07 08:24:15","extension":"png","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":257078,"visible":true,"origin":"","legend":"\u003cp\u003eSchematic diagram of the key points and parameters measured in whole spine frontal position\u003c/p\u003e","description":"","filename":"1.png","url":"https://assets-eu.researchsquare.com/files/rs-6864546/v1/ca7046091591260ed6f61259.png"},{"id":86141001,"identity":"7be6b969-5919-4c7a-bd1e-113b21b6ab41","added_by":"auto","created_at":"2025-07-07 08:24:14","extension":"png","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":333648,"visible":true,"origin":"","legend":"\u003cp\u003eSchematic diagram of key points and parameters measured in whole spine lateral position\u003c/p\u003e","description":"","filename":"2.png","url":"https://assets-eu.researchsquare.com/files/rs-6864546/v1/8acc9e6fd739625d8c40dbbf.png"},{"id":86141008,"identity":"5c7ee888-c270-4bf6-8e8b-f36eca4003d3","added_by":"auto","created_at":"2025-07-07 08:24:14","extension":"png","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":198897,"visible":true,"origin":"","legend":"\u003cp\u003eFlowchart of deep learning model construction for whole spine frontal DR images\u003c/p\u003e","description":"","filename":"3.png","url":"https://assets-eu.researchsquare.com/files/rs-6864546/v1/333bec8d81148b23eacf9c6b.png"},{"id":86141040,"identity":"b5d24791-0d09-482d-949b-06b9b07289e8","added_by":"auto","created_at":"2025-07-07 08:24:16","extension":"png","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":193649,"visible":true,"origin":"","legend":"\u003cp\u003eFlowchart of deep learning model construction for whole spine lateral DR images\u003c/p\u003e","description":"","filename":"4.png","url":"https://assets-eu.researchsquare.com/files/rs-6864546/v1/eec2d7dd692e8dc6a70fc0d2.png"},{"id":86141025,"identity":"d5bb7e77-0dbf-48d0-85b4-71e49a538508","added_by":"auto","created_at":"2025-07-07 08:24:15","extension":"png","order_by":5,"title":"Figure 5","display":"","copyAsset":false,"role":"figure","size":149241,"visible":true,"origin":"","legend":"\u003cp\u003eThe accuracy of whole spine model for detecting different vertebrae and key points\u003c/p\u003e","description":"","filename":"5.png","url":"https://assets-eu.researchsquare.com/files/rs-6864546/v1/0f62951e5269a0e1def5eb43.png"},{"id":86141010,"identity":"dab0d549-8cf2-4b4d-8f46-00e3bfa19c50","added_by":"auto","created_at":"2025-07-07 08:24:14","extension":"png","order_by":6,"title":"Figure 6","display":"","copyAsset":false,"role":"figure","size":151734,"visible":true,"origin":"","legend":"\u003cp\u003ePCK curves of different models for predicting whole spine frontal and lateral vertebrae and key points\u003c/p\u003e","description":"","filename":"6.png","url":"https://assets-eu.researchsquare.com/files/rs-6864546/v1/30baad8d998ce48648c1237f.png"},{"id":86141009,"identity":"650d2a0e-10e8-426b-bfd9-6ce8362b1cdc","added_by":"auto","created_at":"2025-07-07 08:24:14","extension":"png","order_by":7,"title":"Figure 7","display":"","copyAsset":false,"role":"figure","size":8248059,"visible":true,"origin":"","legend":"\u003cp\u003eBland-Altman plots and correlation scatter plots of spino-pelvic parameters in whole spine frontal images\u003c/p\u003e","description":"","filename":"7.png","url":"https://assets-eu.researchsquare.com/files/rs-6864546/v1/408d5a62b4a6f7dc77b1c566.png"},{"id":86140988,"identity":"865ee3f0-047f-4191-a16f-383105304e9d","added_by":"auto","created_at":"2025-07-07 08:24:12","extension":"png","order_by":8,"title":"Figure 8","display":"","copyAsset":false,"role":"figure","size":4740670,"visible":true,"origin":"","legend":"\u003cp\u003eBland-Altman plots and correlation scatter plots of spino-pelvic parameters in whole spine\u003c/p\u003e","description":"","filename":"8.png","url":"https://assets-eu.researchsquare.com/files/rs-6864546/v1/9cd9f2f62629f26dfad3b129.png"},{"id":86143870,"identity":"9c4862be-a17b-40b0-85fd-71330a470df0","added_by":"auto","created_at":"2025-07-07 08:48:28","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":14723866,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-6864546/v1/5ec751a6-aa22-4e3a-8906-f5cc5daae9b9.pdf"},{"id":86141056,"identity":"f437e71e-318d-4615-85e1-e09f9a3aad43","added_by":"auto","created_at":"2025-07-07 08:24:17","extension":"docx","order_by":0,"title":"","display":"","copyAsset":false,"role":"supplement","size":5346228,"visible":true,"origin":"","legend":"","description":"","filename":"supplementalmaterial.docx","url":"https://assets-eu.researchsquare.com/files/rs-6864546/v1/27e405c6a5a7da519b3d16a6.docx"}],"financialInterests":"No competing interests reported.","formattedTitle":"Deep Learning-Based Quantitative Measurement Study of Spino-Pelvic Parameters in Adolescent Idiopathic Scoliosis Patients","fulltext":[{"header":"INTRODUCTION","content":"\u003cp\u003eScoliosis is a complex three-dimensional spinal deformity often characterized by changes in the sagittal and coronal spinal curves, as well as vertebral axis rotation. Adolescent idiopathic scoliosis (AIS) is the most common form of scoliosis, accounting for 75\u0026ndash;80% of cases[\u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e].It is also the most prevalent spinal deformity among adolescents, with a reported prevalence of 1.5\u0026ndash;3% in individuals aged 10\u0026ndash;18 years, occurring more frequently in females[\u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2\u003c/span\u003e].In China, AIS has emerged as the third most prevalent health concern among adolescents, following obesity and myopia[\u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eDue to the age of onset during adolescence, a period of rapid skeletal development, AIS can progress quickly, leading to clinical manifestations such as rib hump, uneven shoulders, leg-length discrepancies, and cardiopulmonary dysfunction. These symptoms severely impact physical appearance, mental health, and quality of life. Early detection and screening of AIS are critical for preventing disease progression and initiating timely treatment strategies[\u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e4\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eCurrently, commonly employed methods for scoliosis screening include visual inspection, the Adams forward flexion test, and trunk rotation angle measurements. Patients who screen positive are referred to orthopedic clinics for whole-spine DR imaging to confirm the diagnosis. DR whole-spine imaging is cost-effective, convenient, and efficient. It not only delineates the morphological and structural characteristics of scoliosis but also provides essential information for monitoring disease progression and formulating diagnostic and treatment plans. As such, it is the most widely used imaging technique for assessing scoliosis severity, monitoring its progression, and evaluating spinal balance and treatment outcomes[\u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e5\u003c/span\u003e]. The diagnostic criterion for scoliosis is a Cobb angle of \u0026ge;\u0026thinsp;10\u0026deg; on standing posterior-anterior whole-spine radiographs. The Scoliosis Research Society (SRS) adopted the Cobb angle as the standard method for quantifying scoliosis deformity in 1966, and it remains the most commonly used measurement index[\u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eSpino-pelvic parameters, which describe the three-dimensional alignment of the spine and pelvis, play a crucial role in the clinical diagnosis, evaluation of scoliosis progression, surgical planning, and prognosis of AIS. Manual measurement methods, such as drawing lines on X-ray films with a ruler or protractor or using angle measurement tools in picture archiving and communication systems (PACS), are labor-intensive, require substantial technical experience, and are prone to intra- and inter-observer errors ranging from 2.8\u0026deg; to 8\u0026deg;[\u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e].Although computer-aided diagnosis (CAD) has improved measurement efficiency through semi-automated methods, manual intervention remains necessary for selecting end vertebrae and determining endplate positions. Variability in endplate selection can result in measurement discrepancies, even among experienced physicians. Moreover, manual operation significantly increases physician workload.\u003c/p\u003e \u003cp\u003eWith advancements in artificial intelligence (AI), deep learning (DL) has emerged as one of the fastest-growing fields within AI. DL technology has been successfully applied to spine image segmentation, disease diagnosis, and parameter measurement[\u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e8\u003c/span\u003e, \u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e9\u003c/span\u003e]. The automated measurement and analysis of spino-pelvic parameters in whole-spine DR images of AIS patients have become a focus of research in DL[\u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e]. However, existing studies have predominantly focused on the automatic measurement of the Cobb angle in posterior-anterior whole-spine DR images,ignoring whole-spine lateral spino-pelvic parameters[\u003cspan additionalcitationids=\"CR12\" citationid=\"CR11\" class=\"CitationRef\"\u003e11\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e13\u003c/span\u003e]. Few studies have utilized biplane X-ray images generated by the EOS system in combination with lateral DR images, and the generalizability of these models remains uncertain[\u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e14\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eTherefore,this study aimed to develope a DL-based tool for the automated measurement of spino-pelvic parameters in AIS patients using two-dimensional frontal and lateral whole-spine DR images and to evaluate its performance and scalability. The results could provide a quantitative, accurate, and efficient automated measurement technique for clinical diagnosis, treatment, and follow-up of AIS patients.\u003c/p\u003e"},{"header":"Materials and Methods","content":"\u003cdiv id=\"Sec3\" class=\"Section2\"\u003e \u003ch2\u003eDatasets\u003c/h2\u003e \u003cp\u003e The study was conducted in accordance with the Declaration of Helsinki (as revised in 2013). The Ethics Committee of Gansu Provincial Hospital of Traditional Chinese Medicine approved this retrospective study (No. 2020-112-01). The requirement for informed consent was waived because only retrospective imaging data were used.\u003c/p\u003e \u003cp\u003eImaging data of adolescent scoliosis patients who underwent whole-spine digital radiography (DR) at Gansu Provincial Hospital of Traditional Chinese Medicine between January 2018 and October 2022 were retrospectively collected.A total of 1,652 whole-spine frontal and lateral DR radiographs from patients with a clinical diagnosis of AIS were included.Inclusion Criteria:(1)Age: 10\u0026ndash;18 years. (2) Whole-spine frontal/lateral radiographs demonstrating the full length of the spine, from the external auditory canal to the femoral head, in patients with a clinical diagnosis of AIS.Exclusion Criteria: (1) Images with significant artifacts or unaligned splices. (2) Incomplete DR images due to positioning issues or deviations from imaging standards. (3) History of spinal surgery or trauma.(4)Presence of metabolic bone disease,spinal tuberculosis,or spinal tumors.(5)Clinical diagnosis of congenital or neuromuscular spinal deformities.\u003c/p\u003e \u003cp\u003eFinally,1,348 whole-spine frontal/lateral DR images were ultimately included, comprising 674 frontal and 674 lateral images. They were randomly divided into training, validation, and test sets based on scoliosis severity in proportions of 65%, 15%, and 20%, respectively. The training and test datasets were completely independent to ensure an objective assessment of the model.\u003c/p\u003e \u003c/div\u003e\n\u003ch3\u003eAnnotation of Key Points in Whole-Spine DR Images and Measurement of Spino-Pelvic Parameters\u003c/h3\u003e \u003cp\u003eTo calculate clinically relevant parameters, the naming of each landmark was standardized. The naming rules for key points are illustrated using the L1 vertebral body as an example. For the whole-spine frontal DR images, the key points are named as follows: L1HR refers to the right margin point of the superior edge of the vertebral body; L1FR refers to the right margin point of the inferior edge of the vertebral body; L1HL refers to the left margin point of the superior edge of the vertebral body; and L1FL refers to the left margin point of the inferior edge of the vertebral body. For the whole-spine lateral DR images, the key points are named as follows: L1HA refers to thefrontal margin point of the superior edge of the vertebral body; L1HP refers to the posterior margin point of the superior edge of the vertebral body; L1FA refers to thefrontal margin point of the inferior edge of the vertebral body; and L1FP refers to the posterior margin point of the inferior edge of the vertebral body. The center of the left femoral head is designated as L-FHCE, while the center of the right femoral head is designated as R-FHCE. Figures\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003e and \u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003e depict all key landmark annotations, their specific names, and the parameter measurements[\u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e15\u003c/span\u003e, \u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e16\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eTwo diagnostic radiologists (R1 and R2), with 10 and 5 years of experience in diagnostic bone and muscle imaging, respectively, were trained in a standardized manner to independently perform pixel-level point labeling of key landmarks in 1,348 whole-spine DR images. This process was conducted using JPHV-specific software designed for model training and validation. To assess intra-observer reliability, the radiologist R2 re-labeled the test set after an interval of six weeks. A third diagnostic radiologist (R3), with 20 years of experience in diagnostic bone and muscle imaging, reviewed all annotations and measurements performed by R1 and R2. R3 corrected any inaccurate annotations and measurements, ensuring the final annotated points and measurements were accurate. These finalized annotations and measurements served as the reference standard for evaluating the performance of the DL in detecting landmarks and measuring parameters.\u003c/p\u003e\n\u003ch3\u003eModel Construction\u003c/h3\u003e\n\u003cdiv id=\"Sec6\" class=\"Section2\"\u003e \u003ch2\u003eDL Modeling of Whole-Spine Frontal and Lateral DR Images in AIS Patients\u003c/h2\u003e \u003cp\u003eIn recent years, through the use of computer vision technology, keypoint localization can help medical experts to analyze image features more accurately and assist in the diagnosis of diseases.Self-calibrated Convolutions Net(SCNet)[\u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e17\u003c/span\u003e]and High-Resolution Net ( HRNet) [\u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e18\u003c/span\u003e]are the most recent networks applied to keypoint localization.\u003c/p\u003e \u003cp\u003eFor the deep learning model of whole spine frontal DR images of AIS patients, however, when SCNet and HRNet were used for single-stage keypoint localization in this study, the portion of the spine in the whole spine image was relatively small, and the effective image area was too small, resulting in low localization accuracy. When using their two-stage methods separately, due to the left-right symmetry of the orthotopic vertebrae, the problem of the opposite prediction of the key point positions of a small number of vertebrae occurs. Therefore, in order to solve the above problems, the Vertebra-Focused Landmark Detection Network (VF-Net)[\u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e19\u003c/span\u003e]was used in our whole-spine frontal model.\u003c/p\u003e \u003cp\u003eFor the deep learning model for whole spine lateral DR images of AIS patients, the same problem of very low localization accuracy exists in this study when using SCNet and HRNet for single-stage keypoint localization. Due to the occlusion of the spine by the shoulder joint, arm, and lungs in the whole spine lateral image, the use of VF-Net causes the problem of misalignment of vertebral center prediction, resulting in low localization accuracy. By experimental comparison, HRNet two-stage method has higher localization accuracy than SCNet two-stage method, single-stage HRNet, and VF-Net. Therefore, in our model, the whole spine lateral DR image model uses two-stage HRNet for key point localization.\u003c/p\u003e \u003c/div\u003e\n\u003ch3\u003e(1) Whole-Spine Frontal DL Model\u003c/h3\u003e\n\u003cp\u003eThe frontal whole-spine DL model employs the VF-Net backbone network, which output comprises three components: Heatmap, Center Offset, and Corner Offset, as illustrated in Figure \u003cspan refid=\"MOESM1\" class=\"InternalRef\"\u003eS1\u003c/span\u003e[\u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e19\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eDuring training, image augmentation techniques, including horizontal flipping, random rotation(\u0026minus;\u0026thinsp;30\u0026deg; to +\u0026thinsp;30\u0026deg;), random scaling, and resizing to 1024 \u0026times; 512 pixels, were applied. The augmented image was then input into the network to produce the center coordinates, center point offsets, and corner point offsets for each vertebra.For inference, the center coordinates and center point offsets were used to compute the actual coordinates of the vertebral centers. Subsequently, the corner point offsets were used to calculate the coordinates of the four corner points of each vertebra based on the vertebral center coordinates.Based on the calculation method in parameter measurement, the relevant spino-pelvic parameters were then calculated from the coordinates of the predicted landmarks, thus enabling automatic measurement.The model construction flowchart is presented in Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003e.\u003c/p\u003e \u003cdiv id=\"Sec8\" class=\"Section2\"\u003e \u003ch2\u003e(2) Whole-Spine Lateral DL Model\u003c/h2\u003e \u003cp\u003eThe modeling of lateral whole-spine images was conducted in two stages using the HRNet network[\u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e20\u003c/span\u003e] for both stages.\u003c/p\u003e \u003cp\u003eStage 1: The key points at the four corner points of each vertebra were predicted, while the left and right femoral head centers were not considered. For each vertebra, the four corner points were grouped to generate a GT heatmap, resulting in a total of 19 heatmaps corresponding to 19 vertebrae. The network output comprised 19 channels, each predicting the key points for one vertebra. During training, image augmentation methods, including horizontal flipping, random rotation, and random scaling, were applied, after which the image was resized to 1024 \u0026times; 512 pixels and input into the network.\u003c/p\u003e \u003cp\u003eStage 2: Two HRNet models were trained. The first HRNet model had 4 output channels for predicting the four corner points of each vertebra. The second HRNet model had 2 output channels for predicting the left and right femoral head centers. The approximate locations of the vertebral corner points were obtained using the first-stage network. Each vertebra was then cropped, and the image was resized to 640 \u0026times; 640 pixels to train the HRNet model with 4 output channels. Similarly, based on the S1 vertebra's position determined by the first-stage network, the region below the S1 vertebra was segmented, resized to 320 \u0026times; 640 pixels, and input into the HRNet model with 2 output channels.\u003c/p\u003e \u003cp\u003eDuring inference, the exact positions of the four vertebral corner points and the center points of the left and right femoral heads were obtained through the two-stage prediction process. These key points were then integrated into a single image to generate the final prediction of key points. The predicted key points were used to measure spino-pelvic parameters. The model construction flowchart is illustrated in Fig.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003e.\u003c/p\u003e \u003c/div\u003e\n\u003ch3\u003eModel Performance Evaluation\u003c/h3\u003e\n\u003cp\u003eThe inter-observer and intra-observer reliability of the landmark annotations were assessed based on the percentage of annotations within landmark-to-landmark distance thresholds of 1 mm, 2 mm, 3 mm, 4 mm, and 5 mm[\u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e21\u003c/span\u003e].Annotations corrected by a senior diagnostic radiologist(R3)were used as the reference standard.The model\u0026rsquo;s performance was evaluated in two dimensions: the accuracy of landmark localization and the consistency of radiological parameter measurements compared with the reference standard on the test set. Specifically, the Percentage of Correct Key Points (PCK) [\u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e22\u003c/span\u003e]metric was employed to evaluate the accuracy of all landmarks predicted by the model.\u003c/p\u003e \u003cp\u003e\u003cimg src=\"data:image/png;base64,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\" width=\"208\" height=\"62\"\u003e\u003c/p\u003e\u003cp\u003eWhere \u003cimg src=\"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAA8AAAAXCAYAAADUUxW8AAAAAXNSR0IArs4c6QAAAARnQU1BAACxjwv8YQUAAAAJcEhZcwAADsMAAA7DAcdvqGQAAADPSURBVDhP7ZOxCoMwFEVvGkVcnNXdr3PzL5wFfyKu7n6BKC5OLg5uzqJoW+WBWJKWCqVQepaXl8shcCHsegcnudA8xa/KRVGgaRrajijbrusaQgjkeQ7HcRDHMSU70pdXIQxDZFkG3/dhWRYlR6SybdsIggCe5237sizbfEQqu64L0zQxTZNSXFEWNs8zndS8bPsZn5ENwwDnHIwx6LpOt0ek8lpSWZZo2xbDMKCqKvR9T+mOVO66DkmSQNM0pGmKcRwRRRGlO////CbfkoEbGM1MKAIQoVYAAAAASUVORK5CYII=\" width=\"15\" height=\"23\"\u003e is an indicator function evaluating to 1 if the location of landmark \u003cimg src=\"data:image/png;base64,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\" width=\"37\" height=\"24\"\u003e predicted by the model is within \u003cimg src=\"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAA0AAAAPCAYAAAA/I0V3AAAAAXNSR0IArs4c6QAAAARnQU1BAACxjwv8YQUAAAAJcEhZcwAADsMAAA7DAcdvqGQAAADHSURBVDhPzZIhDoQwFERnCwcgnABfgQOJRXOL3ocz1KIJB8BiUCQYDAkaAQn5W5af7LLbFZjNPtNkJtNm/u+NDLiI4PMSfx6yDqJtWwzDANd1sa4rhBBI05Rdwx56pSxLCsOQPM/bLyPf9ykIAnYPPkJVVVHf96S1piRJaJ5nMq+xe/B1T0opOI6DPM9ZeWIdxLZtaJoGUkpWzlhDXddhmibEcczKGWtoHEcsywLTB3Vd773ZObCGoihClmUoiuLR651ffVjgDuWMjen78f+6AAAAAElFTkSuQmCC\" width=\"13\" height=\"15\"\u003e mm of the reference standard \u003cimg src=\"data:image/png;base64,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\" width=\"34\" height=\"37\"\u003e, otherwise evaluating to 0.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003eThe spino-pelvic parameters were calculated using the coordinates of the key points, as specified in the parameter measurement part. The intra-class correlation coefficient (ICC), the Pearson correlation coefficient (r), and the mean absolute error(MAE) were calculated between the model prediction and reference standard to evaluate the performance of the model[\u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e23\u003c/span\u003e].An ICC larger than 0.75 was considered to achieve high consistency between the reference standard and the model in evaluating the radiological parameters. |\u003cem\u003er\u003c/em\u003e|\u0026ge;0.7 indicated a high correlation. Moreover,the agreement was assessed between model estimated measurements and reference standard using the mean difference, standard deviation (SD), and 95% limit of agreement graphically represented by the Bland-Altman plots.\u003c/p\u003e\n\u003ch3\u003eNetworks Comparison for Key Point Detection\u003c/h3\u003e\n\u003cp\u003eThe model performance in detecting all landmarks was compared to different networks(VF-Net、one stage HRNet、one stage SCNet、two stage HRNet-HRNet、two stage SCNet-SCNet) by using PCK on the test sets. All networks received the same training data and strategies for a fair comparison.\u003c/p\u003e \u003cdiv id=\"Sec11\" class=\"Section2\"\u003e \u003ch2\u003eStatistical Analysis\u003c/h2\u003e \u003cp\u003eAll statistical analyses were performed using Python (Scipy, Statsmodels, and Pingouin). \u003cem\u003eP\u003c/em\u003e\u0026lt;0.05 was considered statistically different. The PCK metric was employed to evaluate the reliability of manual labeling and the accuracy of the model in detecting key points. Differences between the reference standard and the model-predicted spino-pelvic parameter values were analyzed using paired t-tests. Additional statistical analyses, including the mean absolute error (MAE), Pearson correlation coefficient (r), intra-class correlation coefficient (ICC), and Bland-Altman plots, were conducted to assess the performance of the model in parameter measurement.\u003c/p\u003e \u003c/div\u003e"},{"header":"Results","content":"\u003cdiv id=\"Sec13\" class=\"Section2\"\u003e \u003ch2\u003eGeneral Data Distributions\u003c/h2\u003e \u003cp\u003eA total of 1,348 whole-spine frontal and lateral DR images from 674 adolescent idiopathic scoliosis patients were included in this study for model training, validation, and testing. The dataset comprised 435 images in the training set, 101 in the validation set, and 138 in the testing set. Among all patients, moderate scoliosis was the most prevalent (47.2%), followed by mild scoliosis (37.2%), while severe scoliosis was the least common (15.6%).\u003c/p\u003e \u003cp\u003eThe study included 198 male and 476 female patients, resulting in a male-to-female ratio of approximately 1:2.4. The proportions of female patients in the training, validation, and testing sets were approximately 69.0%, 75.3%, and 72.5%, respectively. The age group of 13–15 years was the most frequent in all datasets, with proportions of 45.3%, 44.6%, and 40.6% in the training, validation, and testing sets, respectively. No significant differences were observed in the gender and age compositions across the datasets (Table\u0026nbsp;\u003cspan refid=\"Tab1\" class=\"InternalRef\"\u003e1\u003c/span\u003e).\u003c/p\u003e \u003cp\u003e \u003c/p\u003e\u003cdiv class=\"gridtable\"\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c6\" colnum=\"6\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c7\" colnum=\"7\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c8\" colnum=\"8\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c9\" colnum=\"9\"\u003e\u003c/div\u003e\u003ctable float=\"Yes\" id=\"Tab1\" border=\"1\"\u003e\u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 1\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eGeneral Information Distribution of Whole Spine DR Imaging Dataset\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e\u003ccolgroup cols=\"9\"\u003e\u003c/colgroup\u003e\u003cthead\u003e\u003ctr\u003e\u003cth align=\"left\" colname=\"c1\" morerows=\"1\" rowspan=\"2\"\u003e \u003cp\u003eScoliosis severity\u003c/p\u003e \u003c/th\u003e\u003cth align=\"left\" colname=\"c2\" morerows=\"1\" rowspan=\"2\"\u003e \u003cp\u003eNo.of images\u003c/p\u003e \u003cp\u003e(sheets)\u003c/p\u003e \u003c/th\u003e\u003cth align=\"left\" colname=\"c3\" morerows=\"1\" rowspan=\"2\"\u003e \u003cp\u003eMale\u003c/p\u003e \u003c/th\u003e\u003cth align=\"left\" colname=\"c4\" morerows=\"1\" rowspan=\"2\"\u003e \u003cp\u003eFemale\u003c/p\u003e \u003c/th\u003e\u003cth align=\"left\" colname=\"c5\" morerows=\"1\" rowspan=\"2\"\u003e \u003cp\u003eAverage CA\u003c/p\u003e \u003cp\u003e(°)\u003c/p\u003e \u003c/th\u003e\u003cth align=\"left\" colname=\"c6\" morerows=\"1\" rowspan=\"2\"\u003e \u003cp\u003eAverage age\u003c/p\u003e \u003cp\u003e(y)\u003c/p\u003e \u003c/th\u003e\u003cth align=\"left\" colspan=\"3\" nameend=\"c9\" namest=\"c7\"\u003e \u003cp\u003eNo.of images in different age ranges(sheets)\u003c/p\u003e \u003c/th\u003e\u003c/tr\u003e\u003ctr\u003e\u003cth align=\"left\" colname=\"c7\"\u003e \u003cp\u003e10–12\u003c/p\u003e \u003c/th\u003e\u003cth align=\"left\" colname=\"c8\"\u003e \u003cp\u003e13–15\u003c/p\u003e \u003c/th\u003e\u003cth align=\"left\" colname=\"c9\"\u003e \u003cp\u003e16–18\u003c/p\u003e \u003c/th\u003e\u003c/tr\u003e\u003c/thead\u003e\u003ctbody\u003e\u003ctr\u003e\u003ctd align=\"left\" colspan=\"9\" nameend=\"c9\" namest=\"c1\"\u003e \u003cp\u003etraining set\u003c/p\u003e \u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003elight\u003c/p\u003e \u003cp\u003e(10°≤CA\u0026lt;20°)\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e164\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e68\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e96\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e14.4\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e13.7\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e58\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003e63\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"left\" colname=\"c9\"\u003e \u003cp\u003e42\u003c/p\u003e \u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003emiddle\u003c/p\u003e \u003cp\u003e(20°≤CA\u0026lt;40°)\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e199\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e46\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e153\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e29.2\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e14.4\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e37\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003e95\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"left\" colname=\"c9\"\u003e \u003cp\u003e67\u003c/p\u003e \u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003esevere(CA ≥ 40°)\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e72\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e21\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e51\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e56.7\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e15.2\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e11\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003e39\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"left\" colname=\"c9\"\u003e \u003cp\u003e22\u003c/p\u003e \u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colspan=\"9\" nameend=\"c9\" namest=\"c1\"\u003e \u003cp\u003evalidation set\u003c/p\u003e \u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003elight\u003c/p\u003e \u003cp\u003e(10°≤CA\u0026lt;20°)\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e31\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e11\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e20\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e14.4\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e13.4\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e13\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003e11\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"left\" colname=\"c9\"\u003e \u003cp\u003e7\u003c/p\u003e \u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003emiddle\u003c/p\u003e \u003cp\u003e(20°≤CA\u0026lt;40°)\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e48\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e9\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e39\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e28.5\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e14.0\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e13\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003e24\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"left\" colname=\"c9\"\u003e \u003cp\u003e11\u003c/p\u003e \u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003esevere(CA ≥ 40°)\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e22\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e5\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e17\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e48.9\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e14.3\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e5\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003e10\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"left\" colname=\"c9\"\u003e \u003cp\u003e7\u003c/p\u003e \u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colspan=\"9\" nameend=\"c9\" namest=\"c1\"\u003e \u003cp\u003etesting set\u003c/p\u003e \u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003elight\u003c/p\u003e \u003cp\u003e(10°≤CA\u0026lt;20°)\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e56\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e20\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e36\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e14.7\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e13.9\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e21\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003e19\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"left\" colname=\"c9\"\u003e \u003cp\u003e16\u003c/p\u003e \u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003emiddle\u003c/p\u003e \u003cp\u003e(20°≤CA\u0026lt;40°)\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e71\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e14\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e57\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e28.0\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e14.0\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e19\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003e34\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"left\" colname=\"c9\"\u003e \u003cp\u003e18\u003c/p\u003e \u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003esevere(CA ≥ 40°)\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e11\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e4\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e7\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e50.3\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e15\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e2\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003e3\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"left\" colname=\"c9\"\u003e \u003cp\u003e6\u003c/p\u003e \u003c/td\u003e\u003c/tr\u003e\u003c/tbody\u003e\u003ctfoot\u003e\u003ctr\u003e\u003ctd colspan=\"9\"\u003eNote: Unless otherwise specified, the data in the table is the number of images. The units in parentheses are the units of measurement data, and the corresponding columns of angle (degree) and average age (y) are the values of measurement data.CA, Cobb angle.\u003c/td\u003e\u003c/tr\u003e\u003c/tfoot\u003e\u003c/table\u003e\u003c/div\u003e \u003cp\u003e\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec14\" class=\"Section2\"\u003e \u003ch2\u003eReliability of Manual Annotation\u003c/h2\u003e \u003cp\u003eThe intra- and inter-observer reliability of the annotations for whole-spinefrontal and lateral DR images, labeled by three radiologists (R1, R2, and R3) with varying levels of experience, are shown in Tables\u0026nbsp;\u003cspan refid=\"Tab2\" class=\"InternalRef\"\u003e2\u003c/span\u003e and \u003cspan refid=\"Tab3\" class=\"InternalRef\"\u003e3\u003c/span\u003e.\u003c/p\u003e \u003cp\u003eFor intra-observer comparisons, the PCK within the 3-mm threshold was 97.6% for whole-spinefrontal DR images and 90.5% for whole-spine lateral DR images. For inter-observer comparisons, the PCK within the 3-mm threshold was as follows: Whole-spine frontal DR images: 88.8% (R1 vs. R2), 98.7% (R1 vs. R3), and 94.0% (R2 vs. R3). Whole-spine lateral DR images: 86.3% (R1 vs. R2), 96.6% (R1 vs. R3), and 91.8% (R2 vs. R3).\u003c/p\u003e \u003cp\u003e \u003c/p\u003e\u003cdiv class=\"gridtable\"\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e\u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e\u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e\u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e\u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e\u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c6\" colnum=\"6\"\u003e\u003c/div\u003e\u003ctable float=\"Yes\" id=\"Tab2\" border=\"1\"\u003e\u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 2\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003ePCK (%) of manual landmark annotation in whole spine frontal DR images\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e\u003ccolgroup cols=\"6\"\u003e\u003c/colgroup\u003e\u003cthead\u003e\u003ctr\u003e\u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eThreshold(mm)\u003c/p\u003e \u003c/th\u003e\u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003e1\u003c/p\u003e \u003c/th\u003e\u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003e2\u003c/p\u003e \u003c/th\u003e\u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003e3\u003c/p\u003e \u003c/th\u003e\u003cth align=\"left\" colname=\"c5\"\u003e \u003cp\u003e4\u003c/p\u003e \u003c/th\u003e\u003cth align=\"left\" colname=\"c6\"\u003e \u003cp\u003e5\u003c/p\u003e \u003c/th\u003e\u003c/tr\u003e\u003c/thead\u003e\u003ctbody\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eintra-observer reliability\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e72.6\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e94.5\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e97.6\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e98.5\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e98.9\u003c/p\u003e \u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003einter-observer reliability\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u0026nbsp;\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eR1 vs R2\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e39.5\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e76.3\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e88.8\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e93.7\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e96.2\u003c/p\u003e \u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eR1 vs R3\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e81.6\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e96.3\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e98.7\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e99.5\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e99.7\u003c/p\u003e \u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eR2 vs R3\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e60.7\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e87.2\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e94.0\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e96.7\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e98.1\u003c/p\u003e \u003c/td\u003e\u003c/tr\u003e\u003c/tbody\u003e\u003c/table\u003e\u003c/div\u003e \u003cp\u003e\u003c/p\u003e \u003cp\u003ePCK, percentage of correct key points.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e\u003cdiv class=\"gridtable\"\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e\u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e\u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e\u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e\u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e\u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c6\" colnum=\"6\"\u003e\u003c/div\u003e\u003ctable float=\"Yes\" id=\"Tab3\" border=\"1\"\u003e\u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 3\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003ePCK (%) of manual landmark annotation in whole spine lateral DR images\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e\u003ccolgroup cols=\"6\"\u003e\u003c/colgroup\u003e\u003cthead\u003e\u003ctr\u003e\u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eThreshold(mm)\u003c/p\u003e \u003c/th\u003e\u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003e1\u003c/p\u003e \u003c/th\u003e\u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003e2\u003c/p\u003e \u003c/th\u003e\u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003e3\u003c/p\u003e \u003c/th\u003e\u003cth align=\"left\" colname=\"c5\"\u003e \u003cp\u003e4\u003c/p\u003e \u003c/th\u003e\u003cth align=\"left\" colname=\"c6\"\u003e \u003cp\u003e5\u003c/p\u003e \u003c/th\u003e\u003c/tr\u003e\u003c/thead\u003e\u003ctbody\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eintra-observer reliability\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e61.7\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e85.3\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e90.5\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e92.8\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e94.1\u003c/p\u003e \u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003einter-observer reliability\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u0026nbsp;\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eR1 vs R2\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e42.0\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e75.9\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e86.3\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e91.1\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e93.3\u003c/p\u003e \u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eR1 vs R3\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e77.4\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e93.3\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e96.6\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e97.7\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e98.1\u003c/p\u003e \u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eR2 vs R3\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e63.0\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e86.1\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e91.8\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e92.1\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e94.2\u003c/p\u003e \u003c/td\u003e\u003c/tr\u003e\u003c/tbody\u003e\u003c/table\u003e\u003c/div\u003e \u003cp\u003e\u003c/p\u003e \u003cp\u003ePCK, percentage of correct key points.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec15\" class=\"Section2\"\u003e \u003ch2\u003eLandmark Detection Performance\u003c/h2\u003e \u003cp\u003eThe model’s performance in detecting key points in whole-spinefrontal and lateral DR images is presented in Tables\u0026nbsp;\u003cspan refid=\"Tab4\" class=\"InternalRef\"\u003e4\u003c/span\u003e and \u003cspan refid=\"Tab5\" class=\"InternalRef\"\u003e5\u003c/span\u003e. Within the 3-mm threshold relative to the reference standard: For the whole-spine frontal model, the PCK for vertebral body key points ranged from 82.4–95.3%; For the whole-spine lateral model, the PCK for vertebral body key points ranged from 82.2–92.4%; The PCK for detecting the center points of the left and right femoral heads was 77.9% and 62.8%, respectively.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e\u003cdiv class=\"gridtable\"\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e\u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e\u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e\u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e\u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e\u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c6\" colnum=\"6\"\u003e\u003c/div\u003e\u003ctable float=\"Yes\" id=\"Tab4\" border=\"1\"\u003e\u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 4\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003ePCK (%) of different vertebral bodies in the whole spine frontal model\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e\u003ccolgroup cols=\"6\"\u003e\u003c/colgroup\u003e\u003cthead\u003e\u003ctr\u003e\u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eThreshold(mm)\u003c/p\u003e \u003c/th\u003e\u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003e1\u003c/p\u003e \u003c/th\u003e\u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003e2\u003c/p\u003e \u003c/th\u003e\u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003e3\u003c/p\u003e \u003c/th\u003e\u003cth align=\"left\" colname=\"c5\"\u003e \u003cp\u003e4\u003c/p\u003e \u003c/th\u003e\u003cth align=\"left\" colname=\"c6\"\u003e \u003cp\u003e5\u003c/p\u003e \u003c/th\u003e\u003c/tr\u003e\u003c/thead\u003e\u003ctbody\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eUV\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e53.1\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e63.5\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e89.2\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e89.2\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e89.2\u003c/p\u003e \u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eLV\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e42.1\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e60.1\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e82.4\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e83.8\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e83.8\u003c/p\u003e \u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eAV\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e55.9\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e70.3\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e87.8\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e87.8\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e87.8\u003c/p\u003e \u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eC7\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e69.7\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e91.0\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e94.5\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e95.9\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e95.9\u003c/p\u003e \u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eT1\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e43.3\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e87.9\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e95.3\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e95.9\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e95.9\u003c/p\u003e \u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eS1\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e22.1\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e67.6\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e93.2\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e94.6\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e96.0\u003c/p\u003e \u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eMean\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e47.7\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e73.4\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e90.4\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e91.2\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e91.4\u003c/p\u003e \u003c/td\u003e\u003c/tr\u003e\u003c/tbody\u003e\u003c/table\u003e\u003c/div\u003e \u003cp\u003e\u003c/p\u003e \u003cp\u003e \u003c/p\u003e\u003cdiv class=\"BlockQuote\"\u003e \u003cp\u003ePCK, Percentage of Correct Key points.UV,upper vertebrae;LV,lower vertebrace;AV,apical vertebrae;C7,7th cervical vertebrae;T1,1st thoracic vertebra;S1,1st sacral vertebra.Mean,The mean Percentage of Correct Key Points.\u003c/p\u003e \u003c/div\u003e \u003cp\u003e\u003c/p\u003e \u003cp\u003e \u003c/p\u003e\u003cdiv class=\"gridtable\"\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e\u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e\u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e\u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e\u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e\u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c6\" colnum=\"6\"\u003e\u003c/div\u003e\u003ctable float=\"Yes\" id=\"Tab5\" border=\"1\"\u003e\u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 5\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003ePCK (%) of different vertebral bodies and key points in the whole spine lateral model\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e\u003ccolgroup cols=\"6\"\u003e\u003c/colgroup\u003e\u003cthead\u003e\u003ctr\u003e\u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eThreshold(mm)\u003c/p\u003e \u003c/th\u003e\u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003e1\u003c/p\u003e \u003c/th\u003e\u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003e2\u003c/p\u003e \u003c/th\u003e\u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003e3\u003c/p\u003e \u003c/th\u003e\u003cth align=\"left\" colname=\"c5\"\u003e \u003cp\u003e4\u003c/p\u003e \u003c/th\u003e\u003cth align=\"left\" colname=\"c6\"\u003e \u003cp\u003e5\u003c/p\u003e \u003c/th\u003e\u003c/tr\u003e\u003c/thead\u003e\u003ctbody\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eC7\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e70.0\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e85.0\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e89.5\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e91.0\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e92.8\u003c/p\u003e \u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eT5\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e43.3\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e67.4\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e82.2\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e89.3\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e93.5\u003c/p\u003e \u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eT10\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e69.3\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e86.2\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e90.5\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e91.6\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e91.7\u003c/p\u003e \u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eT12\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e57.2\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e82.1\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e89.8\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e91.7\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e92.8\u003c/p\u003e \u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eL1\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e64.8\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e86.2\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e90.9\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e91.9\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e93.5\u003c/p\u003e \u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eL2\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e64.1\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e88.3\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e92.4\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e94.3\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e94.5\u003c/p\u003e \u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eS1\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e53.4\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e75.2\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e83.3\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e87.8\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e90.5\u003c/p\u003e \u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eL-FHCE\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e20.7\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e52.4\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e77.9\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e86.9\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e89.0\u003c/p\u003e \u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eR-FHCE\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e15.2\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e45.5\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e62.8\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e73.1\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e79.3\u003c/p\u003e \u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eMean\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e56.7\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e78.7\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e86.5\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e89.6\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e91.4\u003c/p\u003e \u003c/td\u003e\u003c/tr\u003e\u003c/tbody\u003e\u003c/table\u003e\u003c/div\u003e \u003cp\u003e\u003c/p\u003e \u003cp\u003ePCK, Percentage of Correct Key points. C7,7th cervical vertebrae;T5,5th thoracic vertebra;T10,10th thoracic vertebra;T12,12th thoracic vertebra;L1,1st lumbar vertebra;L2,2st lumbar vertebra;S1,1st sacral vertebra;L-FHCE,Left femoral head center point;R-FHCE,Right femoral head center point.Mean,The mean Percentage of Correct Key Points.\u003c/p\u003e \u003cp\u003eThe model demonstrated high accuracy in detecting vertebral and key points in whole-spine frontal and lateral images (Fig.\u0026nbsp;5). The detected key points in the test set closely approximated the reference standard (Figure S2).\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec16\" class=\"Section2\"\u003e \u003ch2\u003eComparison of Networks\u003c/h2\u003e \u003cp\u003eOn the test set, the PCK performance of the proposed models was compared to other DL models.\u003c/p\u003e \u003cp\u003eFor the whole-spine frontal model, the PCK within the 2.5–5 mm threshold was higher for VF-Net (proposed model) than for the one-stage HRNet, one-stage ScNet, two-stage HRNet, and two-stage ScNet. The highest PCKs for each model were as follows: 93.5% (VF-Net), 67.9% (one-stage HRNet), 66.4% (one-stage ScNet), 78.3% (two-stage HRNet), and 79.3% (two-stage ScNet).\u003c/p\u003e \u003cp\u003eFor the whole-spine lateral model, the PCK within the 1–5 mm threshold was highest for the two-stage HRNet (proposed model), followed by the two-stage ScNet, one-stage ScNet, one-stage HRNet, and VF-Net. The highest PCKs for each model were as follows: 88.1% (two-stage HRNet), 85.3% (two-stage ScNet), 83.7% (one-stage ScNet), 86.3% (one-stage HRNet), and 85.9% (VF-Net) (Fig.\u0026nbsp;\u003cspan refid=\"Fig5\" class=\"InternalRef\"\u003e6\u003c/span\u003e).\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec17\" class=\"Section2\"\u003e \u003ch2\u003eModel Measurement Performance\u003c/h2\u003e \u003cp\u003eNo significant differences were observed between the predicted spino-pelvic parameter values of the whole-spine frontal/lateral models and the reference standard (p \u0026gt; 0.05) (Tables\u0026nbsp;\u003cspan refid=\"Tab6\" class=\"InternalRef\"\u003e6\u003c/span\u003e and \u003cspan refid=\"Tab7\" class=\"InternalRef\"\u003e7\u003c/span\u003e). Overall, the model’s predicted spino-pelvic parameter values were consistent and reliable when compared to the reference standard(Whole-spine frontal model: ICC: 0.9–1.0, r: 0.8–1.0, MAE: 1.0–3.7;Whole-spine lateral model: ICC: 0.9–1.0, r: 0.9–1.0, MAE: 1.0–3.5). Bland-Altman plots and regression analyses further illustrated the differences and correlations between the model predictions and the reference standard (Figs.\u0026nbsp;\u003cspan refid=\"Fig6\" class=\"InternalRef\"\u003e7\u003c/span\u003e and \u003cspan refid=\"Fig7\" class=\"InternalRef\"\u003e8\u003c/span\u003e).\u003c/p\u003e \u003cp\u003e \u003c/p\u003e\u003cdiv class=\"gridtable\"\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c6\" colnum=\"6\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c7\" colnum=\"7\"\u003e\u003c/div\u003e\u003ctable float=\"Yes\" id=\"Tab6\" border=\"1\"\u003e\u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 6\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eComparison between the model and the reference standard for the measurement of spino-pelvic parameter(whole spinefrontal)\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e\u003ccolgroup cols=\"7\"\u003e\u003c/colgroup\u003e\u003cthead\u003e\u003ctr\u003e\u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eParameters\u003c/p\u003e \u003c/th\u003e\u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003ereference standard\u003c/p\u003e \u003c/th\u003e\u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003epredicted value\u003c/p\u003e \u003c/th\u003e\u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003eP\u003c/p\u003e \u003c/th\u003e\u003cth align=\"left\" colname=\"c5\"\u003e \u003cp\u003eICC[95%CI]\u003c/p\u003e \u003c/th\u003e\u003cth align=\"left\" colname=\"c6\"\u003e \u003cp\u003er\u003c/p\u003e \u003c/th\u003e\u003cth align=\"left\" colname=\"c7\"\u003e \u003cp\u003eMAE\u003c/p\u003e \u003c/th\u003e\u003c/tr\u003e\u003c/thead\u003e\u003ctbody\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eCVA(mm)\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e12.9 ± 9.0\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e12.6 ± 9.1\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e0.9\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e1.0[0.9,1.0]\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e1.0\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e1.0\u003c/p\u003e \u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eAVT(mm)\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e20.6 ± 15.0\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e20.6 ± 15.1\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e1.0\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e1.0[0.9,1.0]\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e1.0\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e1.1\u003c/p\u003e \u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colspan=\"7\" nameend=\"c7\" namest=\"c1\"\u003e \u003cp\u003eCobb angle(°)\u003c/p\u003e \u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eLight\u003c/p\u003e \u003cp\u003e(10°≤CA\u0026lt;20°)\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e14.6 ± 2.8\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e15.1 ± 3.8\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e0.6\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e0.9[0.8,0.9]\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e0.9\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e1.0\u003c/p\u003e \u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eMiddle\u003c/p\u003e \u003cp\u003e(20°≤CA\u0026lt;40°)\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e27.3 ± 5.3\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e28.0 ± 6.0\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e0.4\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e0.9[0.8,0.9]\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e0.9\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e2.4\u003c/p\u003e \u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eSevere\u003c/p\u003e \u003cp\u003e(CA ≥ 40°)\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e50.3 ± 10.5\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e52.9 ± 14.8\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e0.7\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e0.9[0.6,0.9]\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e0.8\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e3.7\u003c/p\u003e \u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eTotal_CA(°)\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e22.2 ± 9.9\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e22.8 ± 10.2\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e0.6\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e0.9[0.9,1.0]\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e0.9\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e1.8\u003c/p\u003e \u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eT1 TA(°)\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e4.0 ± 4.2\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e3.9 ± 4.7\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e0.8\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e0.9[0.9,1.0]\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e0.9\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e1.3\u003c/p\u003e \u003c/td\u003e\u003c/tr\u003e\u003c/tbody\u003e\u003c/table\u003e\u003c/div\u003e \u003cp\u003e\u003c/p\u003e \u003cp\u003eData are expressed as the mean ± SD.P \u0026lt; 0.05 was considered statistically different.ICC (95% CI), intra-class correlation coefficient (95% confidence interval); r, Pearson correlation coefficient; MAE, the mean absolute error.CVA,coronal vertical axis;AVT,apical vertebral translation;CA,Cobb angle;T1 TA,T1 tital Angle.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e\u003cdiv class=\"gridtable\"\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e\u003cdiv align=\"char\" char=\"±\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e\u003cdiv align=\"char\" char=\"±\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e\u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e\u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e\u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c6\" colnum=\"6\"\u003e\u003c/div\u003e\u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c7\" colnum=\"7\"\u003e\u003c/div\u003e\u003ctable float=\"Yes\" id=\"Tab7\" border=\"1\"\u003e\u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 7\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eComparison between the model and the reference standard for the measurement of spino-pelvic parameter(whole spine lateralization)\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e\u003ccolgroup cols=\"7\"\u003e\u003c/colgroup\u003e\u003cthead\u003e\u003ctr\u003e\u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eParameters\u003c/p\u003e \u003c/th\u003e\u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003ereference standard\u003c/p\u003e \u003c/th\u003e\u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003epredicted value\u003c/p\u003e \u003c/th\u003e\u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003eP\u003c/p\u003e \u003c/th\u003e\u003cth align=\"left\" colname=\"c5\"\u003e \u003cp\u003eICC[95%CI]\u003c/p\u003e \u003c/th\u003e\u003cth align=\"left\" colname=\"c6\"\u003e \u003cp\u003er\u003c/p\u003e \u003c/th\u003e\u003cth align=\"left\" colname=\"c7\"\u003e \u003cp\u003eMAE\u003c/p\u003e \u003c/th\u003e\u003c/tr\u003e\u003c/thead\u003e\u003ctbody\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eTK(°)\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"char\" char=\"±\" colname=\"c2\"\u003e \u003cp\u003e23.5 ± 9.8\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"char\" char=\"±\" colname=\"c3\"\u003e \u003cp\u003e23.3 ± 10.1\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.9\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.9[0.8,0.9]\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e0.9\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e3.5\u003c/p\u003e \u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eTLK(°)\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"char\" char=\"±\" colname=\"c2\"\u003e \u003cp\u003e8.4 ± 6.1\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"char\" char=\"±\" colname=\"c3\"\u003e \u003cp\u003e8.5 ± 6.3\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.9\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.9[0.8,0.9]\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e0.9\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e2.2\u003c/p\u003e \u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eLL(°)\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"char\" char=\"±\" colname=\"c2\"\u003e \u003cp\u003e47.1 ± 10.7\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"char\" char=\"±\" colname=\"c3\"\u003e \u003cp\u003e47.8 ± 11.3\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.6\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e1.0[0.9,1.0]\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e1.0\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e2.5\u003c/p\u003e \u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eSVA(°)\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"char\" char=\"±\" colname=\"c2\"\u003e \u003cp\u003e21.8 ± 16.9\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"char\" char=\"±\" colname=\"c3\"\u003e \u003cp\u003e21.9 ± 17.0\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e1.0\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e1.0[1.0,1.0]\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e1.0\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e1.1\u003c/p\u003e \u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eSS(°)\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"char\" char=\"±\" colname=\"c2\"\u003e \u003cp\u003e40.6 ± 21.0\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"char\" char=\"±\" colname=\"c3\"\u003e \u003cp\u003e40.4 ± 21.2\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e1.0\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e1.0[1.0,1.0]\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e1.0\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e1.5\u003c/p\u003e \u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003ePI(°)\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"char\" char=\"±\" colname=\"c2\"\u003e \u003cp\u003e47.1 ± 10.2\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"char\" char=\"±\" colname=\"c3\"\u003e \u003cp\u003e46.3 ± 10.8\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.6\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e1.0[1.0,1.0]\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e1.0\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e1.9\u003c/p\u003e \u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003ePT(°)\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"char\" char=\"±\" colname=\"c2\"\u003e \u003cp\u003e10.7 ± 6.9\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"char\" char=\"±\" colname=\"c3\"\u003e \u003cp\u003e10.1 ± 6.9\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.5\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e1.0[1.0,1.0]\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e1.0\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e1.0\u003c/p\u003e \u003c/td\u003e\u003c/tr\u003e\u003c/tbody\u003e\u003c/table\u003e\u003c/div\u003e \u003cp\u003e\u003c/p\u003e \u003cp\u003eData are expressed as the mean ± SD.P \u0026lt; 0.05 was considered statistically different.ICC (95% CI), intra-class correlation coefficient (95% confidence interval); r, Pearson correlation coefficient; MAE, the mean absolute error.TK,thoracic kyphosis;TLK,thoracolumbar kyphosis;LL,lumbar lordosis;SVA,sagital vetebare axis;SS,sacral slope;PI,pelvic incidence;PT,pelvic tilt.\u003c/p\u003e \u003c/div\u003e"},{"header":"Discussion","content":"\u003cp\u003e Quantitative, accurate, and efficient measurement of spino-pelvic parameters in adolescent idiopathic scoliosis (AIS) patients is essential for disease diagnosis, treatment planning, and prognosis assessment[24, \u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2\u003c/span\u003e5]. In this study, we developed a deep learning (DL) model for the automatic measurement of spino-pelvic parameters in AIS patients. The model can accurately locate vertebrae and key points on whole-spine frontal and lateral DR images and automatically measure spino-pelvic parameters, achieving predictions comparable to those of senior diagnostic radiology reviewers.\u003c/p\u003e\u003cp\u003eTraditionally, manual measurement of spino-pelvic parameters relies on the subjective judgment and clinical experience of observers, making intra- and inter-observer variations inevitable[\u003cspan citationid=\"CR26\" class=\"CitationRef\"\u003e26\u003c/span\u003e, \u003cspan citationid=\"CR27\" class=\"CitationRef\"\u003e27\u003c/span\u003e]. Errors in the measurement of spino-pelvic parameters have been studied. It has been reported in the literature that the variation in Cobb angle measurements by different observers ranged from 2° to 11°[\u003cspan citationid=\"CR28\" class=\"CitationRef\"\u003e28\u003c/span\u003e].Some scholars also used computer-assisted techniques to measure spondylopelvic parameters in whole spine DR images to study and analyze the differences in manual measurements. The study showed that although computer-assisted semi-automated measurement tools are more advantageous than manual measurement in measuring spino-pelvic parameters[\u003cspan citationid=\"CR27\" class=\"CitationRef\"\u003e27\u003c/span\u003e], the measurement process requires manual selection of vertebrae and marking of anatomical positions for the measurement of relevant parameters, and manual participation is unavoidable. Even when the same diagnostic criteria are followed, experienced diagnostic radiologists may arrive at different assessment results.Recent advances in DL techniques have provided new approaches capable of outperforming professional human operators in tasks such as image classification, target detection, and landmark localization. Several studies have applied DL to scoliosis screening, the prediction of disease progression and brace treatment outcomes, Lenke classification, and the selection of fused vertebral segments[\u003cspan citationid=\"CR29\" class=\"CitationRef\"\u003e29\u003c/span\u003e, \u003cspan citationid=\"CR30\" class=\"CitationRef\"\u003e30\u003c/span\u003e]. However, these studies primarily focused on the automatic measurement of Cobb angle using DL techniques, overlooking other critical spinal frontal planar, sagittal planar, and pelvic parameters.\u003c/p\u003e\u003cp\u003eIn contrast, the DL model developed in this study based on whole-spine frontal and lateral DR images demonstrated the following: (1) The model’s performance in detecting vertebrae key points was comparable to the results of diagnostic radiology reviewers. Within a 3-mm threshold of the reference standard, the model’s PCK for vertebrae key points on frontal and lateral DR images ranged from 82.2–95.3%, while the PCK for predicting the center points of the left and right femoral heads was 77.9% and 62.8%, respectively(Tables\u0026nbsp;\u003cspan refid=\"Tab4\" class=\"InternalRef\"\u003e4\u003c/span\u003e and \u003cspan refid=\"Tab5\" class=\"InternalRef\"\u003e5\u003c/span\u003e;Figure 5). (2) The spino-pelvic parameters measured by the DL model on frontal and lateral DR images exhibited high correlation and consistency with the reference standards (Tables\u0026nbsp;\u003cspan refid=\"Tab6\" class=\"InternalRef\"\u003e6\u003c/span\u003e and \u003cspan refid=\"Tab7\" class=\"InternalRef\"\u003e7\u003c/span\u003e; Figs.\u0026nbsp;\u003cspan refid=\"Fig6\" class=\"InternalRef\"\u003e7\u003c/span\u003e and \u003cspan refid=\"Fig7\" class=\"InternalRef\"\u003e8\u003c/span\u003e). (3) For whole-spine frontal DR images, the VF-Net used in this study demonstrated superior accuracy in detecting vertebrae key points within the 2.5–5 mm threshold compared to SCNet and HRNet. For whole-spine lateral DR images, the two-stage HRNet method proposed in this study outperformed SCNet, VF-Net, and single-stage HRNet in detecting vertebrae and key points within the 1–5 mm threshold (Fig.\u0026nbsp;\u003cspan refid=\"Fig5\" class=\"InternalRef\"\u003e6\u003c/span\u003e).\u003c/p\u003e\u003cp\u003eTo ensure accurate parameter measurements, key points were identified based on corrections provided by senior diagnostic radiology reviewers, which served as the reference standard. Chen et al.[\u003cspan citationid=\"CR31\" class=\"CitationRef\"\u003e31\u003c/span\u003e]reported that data with a mean inter-observer landmark distance within 3 mm is suitable for clinical analysis. Consistent with this, the results of our study showed that most inter-observer distances fell within the 3-mm threshold, confirming the reliability of the labeling and auditing results from the three radiologists (Tables\u0026nbsp;\u003cspan refid=\"Tab2\" class=\"InternalRef\"\u003e2\u003c/span\u003e and \u003cspan refid=\"Tab3\" class=\"InternalRef\"\u003e3\u003c/span\u003e).Furthermore, within the 3-mm threshold, the model-predicted PCK values for the upper and lower end vertebrae and the apical vertebrae were 89.2%, 82.4%, and 87.8%, respectively, for whole-spine frontal DR images, which were comparable to and in some cases slightly higher than the results reported in existing studies[\u003cspan citationid=\"CR32\" class=\"CitationRef\"\u003e32\u003c/span\u003e]. Additionally, the DL model performed well in detecting the T1, C7, and S1 vertebrae, achieving PCK values of 95.3%, 94.5%, and 93.2%, respectively, within the 3-mm threshold. For lateral DR images, the model’s accuracy in detecting vertebrae and key points was slightly lower than that for frontal DR images, with mean PCK values of 90.4% and 86.5%, respectively. This difference can be attributed to challenges such as the overlap of lung tissue and thoracic bony structures, air-filled abdominal intestinal tubes, and pelvic anatomical structures in lateral images, which hinder precise identification of vertebrae key points.The model achieved the lowest PCK values for the femoral head centers, followed by the T5 vertebra (PCK: 82.2%). This finding aligns with previous studies[\u003cspan citationid=\"CR33\" class=\"CitationRef\"\u003e33\u003c/span\u003e]and is primarily due to the lower image contrast in the pelvic region of lateral DR images, which impacts the detection of femoral head centers. Additionally, anatomical interference from structures such as the scapula, clavicle, costovertebral joints, aorta, and mediastinum contributes to boundary blurring of the T5 vertebra, posing challenges even for diagnostic radiologists. These limitations are objective, unavoidable, and independent of the DL model itself.\u003c/p\u003e\u003cp\u003eSeveral studies have explored DL techniques for automated spinal curvature measurements in AIS patients, achieving promising results. For example, Horng et al.[\u003cspan citationid=\"CR34\" class=\"CitationRef\"\u003e34\u003c/span\u003e] validated their model’s performance within a Cobb angle range of 0–20.1° but did not evaluate it for Cobb angles exceeding 20.1°. Similarly, Zhang et al.[\u003cspan citationid=\"CR35\" class=\"CitationRef\"\u003e35\u003c/span\u003e]developed a vertebral angle-point detection method for Cobb angle measurement but did not employ statistical methods to assess its consistency with manual measurements. Wu et al. [\u003cspan citationid=\"CR36\" class=\"CitationRef\"\u003e36\u003c/span\u003e] proposed a Multi-View Correlation Network (MVC-Net) for fully automated spinal curvature assessment, achieving a mean absolute error (MAE) of 4.0° for the Cobb angle. Sardjono et al.[\u003cspan citationid=\"CR37\" class=\"CitationRef\"\u003e37\u003c/span\u003e]tested a DL model for automated Cobb angle measurement, reporting an MAE of 3.9°. The clinically acceptable error range for automated Cobb angle measurements is 3–5° [\u003cspan citationid=\"CR38\" class=\"CitationRef\"\u003e38\u003c/span\u003e].\u003c/p\u003e\u003cp\u003eCompared to these studies, the DL model developed in this study demonstrated superior performance, particularly in predicting severe Cobb angles. The MAEs for mild, moderate, severe, and overall Cobb angles were 1.0°, 2.4°, 3.7°, and 1.8°, respectively, which are lower than previously reported values. Huang et al.[\u003cspan citationid=\"CR39\" class=\"CitationRef\"\u003e39\u003c/span\u003e]applied a convolutional neural network (CNN) to 300 whole-spine frontal DR images of AIS patients, reporting intra-class correlation coefficients (ICCs) of 0.6, 0.9, and 0.8 for mild, moderate, and severe Cobb angles, respectively. In contrast, the ICCs of mild, moderate, severe, and overall Cobb angles for the DL model in this study were same (ICC = 0.9), indicating higher reliability and accuracy.Moreover, the model demonstrated excellent performance in predicting coronal vertical axis (CVA) and apical vertebral translation (AVT) parameters, with MAEs of 1.0 mm and 1.1mm, respectively, and ICCs of 1.0 for both parameters, comparable to the results reported by Wu et al. [\u003cspan citationid=\"CR32\" class=\"CitationRef\"\u003e32\u003c/span\u003e]for multi-stage integrated network models. Additionally, T1 tilt angle measurements were included in this study, showing high accuracy with an MAE of 1.3°, an ICC of 0.9, and a Pearson correlation coefficient (r) of 0.9, approaching the results obtained by senior diagnostic radiology reviewers.\u003c/p\u003e\u003cp\u003eCombining coronal and sagittal measurements of spino-pelvic parameters to provide a comprehensive analysis at the AIS patient level is a notable advantage. Zerouali et al. [\u003cspan citationid=\"CR40\" class=\"CitationRef\"\u003e40\u003c/span\u003e]developed a DL model for the automatic assessment of spino-pelvic coronal and sagittal parameters based on data from 100 pediatric and adult scoliosis patients. The model demonstrated good consistency and accuracy (ICC ≥ 0.9; MAE ≤ 4.4° or 2.7 mm) for most parameters, except for thoracic kyphosis (TK), where the ICC was 0.58 and the MAE was 8.7°. Wu et al.[\u003cspan citationid=\"CR36\" class=\"CitationRef\"\u003e36\u003c/span\u003e]proposed the MVC-Net network architecture, which overcomes the challenges of sternal overlap in whole-spine lateral images. Their model achieved a mean absolute error (MAE) of 4.1° in predicting the lateral Cobb angle. Similarly, although the performance of our DL model in predicting TK for whole-spine lateral images was slightly inferior to that for other parameters, consistent with existing studies, the MAE for TK was 3.5°, with ICC and Pearson correlation coefficient (r) values of 0.9. This indicates that the error is smaller, and the consistency and correlation are higher than those reported for the aforementioned DL models. In addition, Galbusera et al.[\u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e14\u003c/span\u003e]developed a DL model using biplanar radiographs acquired via the EOS imaging system, with a standard deviation range of 2.7° (Pelvic Tilt, PT) to 11.5° (L1-L5 spinalfrontal convexity angle) for predicting sagittal spino-pelvic parameters. By comparison, the DL model developed in the present study achieved a standard deviation range of 1.7° (Pelvic Tilt,PT) to 4.6° (thoracic kyphosis,TK), highlighting its superior performance in measuring sagittal spino-pelvic parameters in AIS patients.\u003c/p\u003e\u003cp\u003eWith the advancement and optimization of DL network models, SCNet and HRNet, which are state-of-the-art methods for detecting key points in human posture estimation tasks, exhibit low accuracy when locating vertebrae in whole-spine frontal DR images using single-stage approaches. Furthermore, incorrect key point detection occurs when their two-stage methods are applied separately. To address these limitations, the present study adopted the VF-Net single-stage approach for the whole-spine frontal DL model, incorporating a ResNet34 network as the encoder. This strategy improves the detection accuracy of vertebrae and key points by employing vertebrae center heatmaps, center offsets,and corner offsets. Fully automated detection of vertebral key points demonstrated greater accuracy, particularly within the range of 2.5-5 mm. Based on the accurate identification of vertebrae key points, any two vertebrae from the 17 included (12 thoracic and 5 lumbar) were selected to calculate the angle between the upper endplate of the superior vertebra and the lower endplate of the inferior vertebra. The largest angle obtained was identified as the Cobb angle, and the corresponding vertebrae were labeled as the upper and lower end vertebrae[\u003cspan citationid=\"CR41\" class=\"CitationRef\"\u003e41\u003c/span\u003e]. Furthermore, the distance between the center of each vertebra and the center sacral vertical line within the curvature range was calculated, with the largest distance defined as the apical vertebral translation (AVT) and the corresponding vertebra identified as the apical vertebra. Additionally, the model automatically measured other spino-pelvic parameters of the whole spine in the frontal plane.\u003c/p\u003e\u003cp\u003eInter-model comparisons revealed that the two-stage HRNet model demonstrated superior performance in detecting key points in whole-spine lateral images compared to other models. In this study, the two-stage HRNet method, combined with end-to-end learning, effectively maintained high resolution, enabling the automatic localization of vertebral body corner points and bilateral femoral head centers in whole-spine lateral images.The model exhibited a significant advantage over SCNet, VF-Net, and single-stage HRNet in detecting vertebral bodies and key points within a threshold range of 1–5 mm. This facilitated the automatic measurement of spino-pelvic parameters in the lateral plane. Although the performance of the whole-spine lateral model for detecting bilateral femoral head centers at the 3-mm threshold was lower than that for vertebrae detection, this discrepancy had minimal impact on pelvic parameter measurements. Specifically, the distance between the midpoints of the bilateral femoral head center lines and the midpoint of the first sacral upper endplate was approximately 15 cm, reducing the influence of center point detection errors on pelvic parameter assessments.\u003c/p\u003e\u003cp\u003eThere are several limitations to this study. First, this is a single-center retrospective study, which lacks an external validation dataset. Variations in image quality may limit the performance of the DL model. Future work will focus on developing DL models using multicenter databases to address this limitation. Second, the current DL model may not accurately predict spino-pelvic parameters in patients with vertebral sequence abnormalities, such as lumbosacral transitional vertebrae or congenital vertebral anomalies. The accuracy of our DL method in predicting landmarks relies heavily on the correct identification of vertebrae and their boundary corner points. To address this issue, it may be necessary to design an algorithm that determines the number of vertebrae before landmark localization, while simultaneously increasing the diversity of scoliosis cases in the training dataset. Third, the current study includes a predominance of AIS patients with moderate scoliosis, while the number of patients with severe scoliosis remains relatively small. Consequently, the training dataset may not fully represent complex clinical settings. Future studies will aim to incorporate a broader range of cases to improve model generalizability.\u003c/p\u003e\u003cp\u003eIn conclusion, this study developed and validated an automated DL-based measurement model for spino-pelvic parameters on whole-spine frontal and lateral DR images in AIS patients. The model demonstrated performance comparable to senior diagnostic radiology reviewers in detecting vertebrae, identifying key points, and measuring most spino-pelvic parameters. Furthermore, its performance was close to or exceeded that of previously reported DL models. In future research, we plan to conduct multicenter studies and collect external validation datasets encompassing a wide range of scoliosis types and severities. This will enhance the robustness of the model and provide a reliable, efficient tool for intelligent assessment in clinical practice, optimizing the diagnostic and therapeutic workflow for AIS patients.\u003c/p\u003e"},{"header":"Declarations","content":"\u003cp\u003e\u003cstrong\u003eAcknowledgments\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThis work was supported by the General Project of the Joint Research Fund of Gansu Province (23JRRA1542),Gansu Provincial Hospital;National Natural Science Foundation of China (NSFC)(82360358),Gansu Provincial Hospital.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eFootnote\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e\u003cem\u003eConflicts Of Interest\u003c/em\u003e\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eAll authors have completed the ICMJE uniform disclosure form.\u0026nbsp;Guohua Cheng is a consultant of Hangzhou Jianpei Technology Co., Ltd.Linyang He is an employee of Hangzhou Jianpei Technology Co., Ltd.The authors have no conflicts of interest to declare.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e\u003cem\u003eEthical Statement\u003c/em\u003e\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe authors are accountable for all aspects of the work in ensuring that questions related to the accuracy or integrity of any part of the work are appropriately investigated and resolved. The study was conducted in accordance with the Declaration of Helsinki (as revised in 2013). The study was approved by Gansu Provincial Hospital of Traditional Chinese Medicine approved this retrospective study (No. 2020-112-01). The requirement for informed consent was waived because retrospective imaging data were used.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e\u003cem\u003eHuman Ethics and Consent to Participate declarations\u003c/em\u003e\u003c/strong\u003e: not applicable\u003c/p\u003e\u003ch2\u003eAuthor Contribution\u003c/h2\u003e\u003cp\u003eContributions: (I) Conception and design: Z Chen; (II) Administrative support: D Zhu; (III) Provision of study materials or patients: J Qian; (IV) Collection and assembly of data: W Wang; (V) Data analysis and interpretation: G Cheng; (VI) Manuscript writing: All authors; (VII) Final approval of manuscript: All authors.\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\n\u003cli\u003eKuznia AL, Hernandez AK, Lee LU (2020) Adolescent Idiopathic Scoliosis: Common Questions and Answers. AM FAM PHYSICIAN 101:19-23\u003c/li\u003e\n\u003cli\u003eHorne JP, Flannery R, Usman S (2014) Adolescent idiopathic scoliosis: diagnosis and management. AM FAM PHYSICIAN 89:193\u003c/li\u003e\n\u003cli\u003eLv Z, Lv W, Wang L, Ou J (2023) Development and validation of machine learning-based models for prediction of adolescent idiopathic scoliosis: A retrospective study. MEDICINE 102:e33441. 10.1097/MD.0000000000033441*10.1097/MD.0000000000033441\u003c/li\u003e\n\u003cli\u003eAltaf F, Gibson A, Dannawi Z, Noordeen H (2013) Adolescent idiopathic scoliosis. BMJ-BRIT MED J 346:f2508. 10.1136/bmj.f2508*10.1136/bmj.f2508\u003c/li\u003e\n\u003cli\u003eGarc\u0026iacute;a-Cano E, Ar\u0026aacute;mbula Cos\u0026iacute;o F, Duong L, Bellefleur C, Roy-Beaudry M, Joncas J, Parent S, Labelle H (2018) Prediction of spinal curve progression in Adolescent Idiopathic Scoliosis using Random Forest regression. COMPUT BIOL MED 103:34-43. 10.1016/j.compbiomed.2018.09.029*10.1016/j.compbiomed.2018.09.029\u003c/li\u003e\n\u003cli\u003eThompson GH (2010) Scoliosis Research Society. SPINE 35:134-137\u003c/li\u003e\n\u003cli\u003eZhou Z, Zhu J, Yao C (2023) Vertebral Center Points Locating and Cobb Angle Measurement Based on Deep Learning. Applied Sciences 13:3817. 10.3390/app13063817*10.3390/app13063817\u003c/li\u003e\n\u003cli\u003eGaonkar B, Villaroman D, Beckett J, Ahn C, Macyszyn L (2019) Quantitative Analysis of Spinal Canal Areas in the Lumbar Spine: An Imaging Informatics and Machine Learning Study. AM J NEURORADIOL 40:1586-1591\u003c/li\u003e\n\u003cli\u003eWimmer M, Major D, Novikov AA, B\u0026uuml;hler K (2018) Fully automatic cross-modality localization and labeling of vertebral bodies and intervertebral discs in 3D spinal images. International Journal of Computer Assisted Radiology \u0026amp; Surgery:1-13\u003c/li\u003e\n\u003cli\u003eSan Rom\u0026aacute;n Gaitero A, Shoykhet A, Spyrou I, Stoorvogel M, Vermeer L, Schl\u0026ouml;sser TPC (2023) Imaging Methods to Quantify the Chest and Trunk Deformation in Adolescent Idiopathic Scoliosis: A Literature Review. HEALTHCARE-BASEL 11:1489. 10.3390/healthcare11101489*10.3390/healthcare11101489\u003c/li\u003e\n\u003cli\u003eKokabu T, Kanai S, Kawakami N, Uno K, Kotani T, Suzuki T, Tachi H, Abe Y, Iwasaki N, Sudo H (2021) An algorithm for using deep learning convolutional neural networks with three dimensional depth sensor imaging in scoliosis detection. Spine J 21:980-987. 10.1016/j.spinee.2021.01.022*10.1016/j.spinee.2021.01.022\u003c/li\u003e\n\u003cli\u003eChen B, Xu Q, Wang L, Leung S, Chung J, Li S (2019) An Automated and Accurate Spine Curve Analysis System. IEEE ACCESS 7:124596-124605. 10.1109/ACCESS.2019.2938402*10.1109/ACCESS.2019.2938402\u003c/li\u003e\n\u003cli\u003eDubost F, Collery B, Renaudier A, Roc A (2020) Automated Estimation of the SpinalCurvature via Spine Centerline Extraction with Ensembles of Cascaded Neural Networks. Computational Methods and Clinical Applications for Spine Imaging.CSI 2019. Lecture Notes in Computer Science 11963:88-94. 10.1007/978-3-030-39752-4_10*10.1007/978-3-030-39752-4_10\u003c/li\u003e\n\u003cli\u003eGalbusera F, Niemeyer F, Wilke HJ, Bassani T, Casaroli G, Anania C, Costa F, Brayda-Bruno M, Sconfienza LM (2019) Fully automated radiological analysis of spinal disorders and deformities: a deep learning approach. EUR SPINE J 28:951-960. 10.1007/s00586-019-05944-z*10.1007/s00586-019-05944-z\u003c/li\u003e\n\u003cli\u003eSingh R, Yadav SK, Sood S, Yadav RK, Rohilla R (2018) Spino-pelvic radiological parameters in normal Indian population. SICOT-J 4:14\u003c/li\u003e\n\u003cli\u003eYao J, Dong B, Sun J, Liu J, Zhang J (2020) Accuracy and Reliability of Computer゛ided Anatomical Measurements for Vertebral Body and Disc Based on Computed Tomography Scans. ORTHOP SURG \u003c/li\u003e\n\u003cli\u003eLee CY, Xie S, Gallagher P, Zhang Z, Tu Z (2014) Deeply-Supervised Nets. Eprint Arxiv:562-570. 10.48550/arXiv.1409.5185*10.48550/arXiv.1409.5185\u003c/li\u003e\n\u003cli\u003eWang J, Sun K, Cheng T, Jiang B, Deng C, Zhao Y, Liu D, Mu Y, Tan M, Wang X, Liu W, Xiao B (2021) Deep High-Resolution Representation Learning for Visual Recognition. IEEE T PATTERN ANAL 43:3349-3364. 10.1109/TPAMI.2020.2983686*10.1109/TPAMI.2020.2983686\u003c/li\u003e\n\u003cli\u003eYi JR, Wu PX, Huang QY, Hui Q (2020) VERTEBRA-FOCUSED LANDMARK DETECTION FOR SCOLIOSIS ASSESSMENT. In. Rutgers University, Piscataway.\u003c/li\u003e\n\u003cli\u003eSun K, Xiao B, Liu D, Wang J (2019) Deep High-Resolution Representation Learning for Human Pose Estimation. In: 2019 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). Cornell University Library.\u003c/li\u003e\n\u003cli\u003ePayer C, Tern D, Bischof H, Urschler M (2019) Integrating Spatial Configuration into Heatmap Regression Based CNNs for Landmark Localization. MED IMAGE ANAL \u003c/li\u003e\n\u003cli\u003eLi T, Wang Y, Qu Y, Dong R, Kang M, Zhao J (2022) Feasibility study of hallux valgus measurement with a deep convolutional neural network based on landmark detection. SKELETAL RADIOL 51:1235-1247\u003c/li\u003e\n\u003cli\u003eYang W, Ye Q, Ming S, Hu X, Jiang Z, Shen Q, He L, Gong X (2020) Feasibility of automatic measurements of hip joints based on pelvic radiography and a deep learning algorithm. EUR J RADIOL 132:109303. 10.1016/j.ejrad.2020.109303*10.1016/j.ejrad.2020.109303\u003c/li\u003e\n\u003cli\u003eBernstein P, Metzler J, Weinzierl M, Seifert C, Kisel W, Wacker M (2021) Radiographic scoliosis angle estimation: spline-based measurement reveals superior reliability compared to traditional COBB method. EUR SPINE J 30:676-685. 10.1007/s00586-020-06577-3*10.1007/s00586-020-06577-3\u003c/li\u003e\n\u003cli\u003eSoffer S, Ben-Cohen A, Shimon O, Amitai MM, Greenspan H, Klang E (2019) Convolutional Neural Networks for Radiologic Images: A Radiologist\u0026apos;s Guide. RADIOLOGY 290:590-606. 10.1148/radiol.2018180547*10.1148/radiol.2018180547\u003c/li\u003e\n\u003cli\u003ePrestigiacomo FG, Hulsbosch MHHM, Bruls VEJ, Nieuwenhuis JJ (2022) Intra- and inter-observer reliability of Cobb angle measurements in patients with adolescent idiopathic scoliosis. SPINE DEFORM 10:79-86. 10.1007/s43390-021-00398-0*10.1007/s43390-021-00398-0\u003c/li\u003e\n\u003cli\u003eWu W, Liang J, Du Y, Tan X, Xiang X, Wang W, Ru N, Le J (2014) Reliability and reproducibility analysis of the Cobb angle and assessing sagittal plane by computer-assisted and manual measurement tools. BMC MUSCULOSKEL DIS 15:33. 10.1186/1471-2474-15-33*10.1186/1471-2474-15-33\u003c/li\u003e\n\u003cli\u003eCheung J, Wever DJ, Veldhuizen AG, Klein JP, Verdonck B, Nijlunsing R, Cool JC, Van Horn JR (2002) The reliability of quantitative analysis on digital images of the scoliotic spine. European spine journal 11:535-542\u003c/li\u003e\n\u003cli\u003eYang J, Zhang K, Fan H, Huang Z, Xiang Y, Yang J, He L, Zhang L, Yang Y, Li R (2019) Development and validation of deep learning algorithms for scoliosis screening using back images. COMMUN BIOL 2:390\u003c/li\u003e\n\u003cli\u003ePhan P, Mezghani N, Wai EK, Guise JD (2013) Artificial neural networks assessing adolescent idiopathic scoliosis: comparison with Lenke classification - ScienceDirect. Spine J 13:1527-1533. 10.1016/j.spinee.2013.07.449.*10.1016/j.spinee.2013.07.449.\u003c/li\u003e\n\u003cli\u003eChen H, Lin C, Wu C, Wang C, Sun Y (2010) Automatic Insall\u0026ndash;Salvati ratio measurement on lateral knee x-ray images using model-guided landmark localization. PHYS MED BIOL 55:6785-6800. 10.1088/0031-9155/55/22/012*10.1088/0031-9155/55/22/012\u003c/li\u003e\n\u003cli\u003eWu C, Meng G, Lian J, Xu J, Gao M, Huang C, Zhang S, Zhang Y, Yu Y, Wang H, He Y, Li X, Li Z (2022) A multi-stage ensemble network system to diagnose adolescent idiopathic scoliosis. EUR RADIOL. 10.1007/s00330-022-08692-9*10.1007/s00330-022-08692-9\u003c/li\u003e\n\u003cli\u003eYeh Y, Weng C, Huang Y, Fu C, Tsai T, Yeh C (2021) Deep learning approach for automatic landmark detection and alignment analysis in whole-spine lateral radiographs. SCI REP-UK 11. 10.1038/s41598-021-87141-x*10.1038/s41598-021-87141-x\u003c/li\u003e\n\u003cli\u003eHorng M, Kuok C, Fu M, Lin C, Sun Y (2019) Cobb Angle Measurement of Spine from X-Ray Images Using Convolutional Neural Network. COMPUT MATH METHOD M 2019:1-18. 10.1155/2019/6357171*10.1155/2019/6357171\u003c/li\u003e\n\u003cli\u003eZhang K, Xu N, Yang G, Wu J (2019) An Automated Cobb Angle Estimation Method Using Convolutional Neural Network with Area Limitation. In: Medical Image Computing and Computer-Assisted Intervention-MICCAI 2019. Springer International Publishing, Switzerland. pp. 775-783.\u003c/li\u003e\n\u003cli\u003eWu H, Bailey C, Rasoulinejad P, Li S (2018) Automated comprehensive Adolescent Idiopathic Scoliosis assessment using MVC-Net. MED IMAGE ANAL 48:1-11. https://doi.org/10.1016/j.media.2018.05.005*https://doi.org/10.1016/j.media.2018.05.005\u003c/li\u003e\n\u003cli\u003eSardjono TA, Wilkinson MH, Veldhuizen AG, van Ooijen PM, Purnama KE, GJ. V (2013) Automatic Cobb Angle Determination From Radiographic Images. SPINE 38:E1256-E1262\u003c/li\u003e\n\u003cli\u003eLechner R, Putzer D, Dammerer D, Liebensteiner M, Bach C, Thaler M (2017) Comparison of two- and three-dimensional measurement of the Cobb angle in scoliosis. INT ORTHOP 41:957-962. 10.1007/s00264-016-3359-0*10.1007/s00264-016-3359-0\u003c/li\u003e\n\u003cli\u003eHuang X, Luo M, Liu L, Wu D, You X, Deng Z, Xiu P, Yang X, Zhou C, Feng G, Wang L, Zhou Z, Fan J, He M, Gao Z, Pu L, Wu Z, Zhou Z, Song Y, Huang S (2022) The Comparison of Convolutional Neural Networks and the Manual Measurement of Cobb Angle in Adolescent Idiopathic Scoliosis. GLOB SPINE J 0:1-10. 10.1177/21925682221098672*10.1177/21925682221098672\u003c/li\u003e\n\u003cli\u003eZerouali M, Parpaleix A, Benbakoura M, Rigault C, Champsaur P, Guenoun D (2023) Automatic deep learning-based assessment of spinopelvic coronal and sagittal alignment. DIAGN INTERV IMAG 15:1-8. https://doi.org/10.1016/j.diii.2023.03.003*https://doi.org/10.1016/j.diii.2023.03.003\u003c/li\u003e\n\u003cli\u003eSun Y, Xing YZ, Zhao ZA, Meng X, Xu G, Hai Y (2022) Comparison of manual versus automated measurement of Cobb angle in idiopathic scoliosis based on a deep learning keypoint detection technology. EUR SPINE J 31:1969-1978. 10.1007/s00586-021-07025-6*10.1007/s00586-021-07025-6\u003c/li\u003e\n\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"hideJournal":false,"highlight":"","institution":"","isAcceptedByJournal":false,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"bmc-medical-imaging","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"bmim","sideBox":"Learn more about [BMC Medical Imaging](http://bmcmedimaging.biomedcentral.com/)","snPcode":"","submissionUrl":"https://www.editorialmanager.com/bmim/default.aspx","title":"BMC Medical Imaging","twitterHandle":"BMC_series","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"em","reportingPortfolio":"BMC Series","inReviewEnabled":true,"inReviewRevisionsEnabled":true},"keywords":"adolescent idiopathic scoliosis, deep learning, whole-spine DR image, spino-pelvic parameters","lastPublishedDoi":"10.21203/rs.3.rs-6864546/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-6864546/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003ch2\u003eObjective\u003c/h2\u003e \u003cp\u003eTo develop a deep learning (DL) model for the automated measurement of spino-pelvic parameters on whole-spine digital radiography (DR) images of adolescent idiopathic scoliosis (AIS) patients and to evaluate its performance and generalizability.\u003c/p\u003e\u003ch2\u003eMethods\u003c/h2\u003e \u003cp\u003eA total of 1348 whole-spine frontal/lateral DR images of AIS patients were collected to train the DL model for predicting fourteen spino-pelvic parameters. The reliability of manual annotation and the model's performance in detecting key points were assessed using Percentage of Correct Keypoints (PCK) values. Differences between the model-predicted parameter values and manual measurements, used as the reference standard, were analyzed using paired t-tests. Further performance evaluations included the mean absolute error (MAE), Pearson correlation coefficient (r), intra-class correlation coefficient (ICC), Bland-Altman plots.\u003c/p\u003e\u003ch2\u003eResults\u003c/h2\u003e \u003cp\u003eThe DL model demonstrated the ability to automatically detect vertebral bodies and key points. The PCK for detecting vertebral body key points in whole-spine DR images ranged from 82.2\u0026ndash;95.3% within a 3-mm threshold. Additionally, the PCK for the left and right femoral heads in whole-spine lateral DR images was 77.9% and 62.3%, respectively. The spino-pelvic parameter values measured by the DL model exhibited a high correlation and agreement with the reference standard (ICC: 0.9\u0026ndash;1.0, r: 0.8\u0026ndash;1.0, MAE: 1.0\u0026ndash;3.7).The Whole frontal model used VF-Net outperformed other networks (one stage HRNet、one stage SCNet、two stage HRNet-HRNet、two stage SCNet-SCNet) in predicting landmarks within a distance threshold of 2.5 to 5 mm;The Whole lateral model used two stage HRNet-HRNet outperformed other networks (VF-Net,one stage HRNet、one stage SCNet、two stage SCNet-SCNet) in predicting landmarks within a distance threshold of 1 to 5 mm.\u003c/p\u003e\u003ch2\u003eConclusions\u003c/h2\u003e \u003cp\u003eThe DL model developed in this study can automatically measure spino-pelvic parameters with performance comparable to that of radiologists. This model is expected to provide an automated measurement tool for clinical practice, thereby improving efficiency in diagnosing, monitoring, and treating AIS patients.\u003c/p\u003e","manuscriptTitle":"Deep Learning-Based Quantitative Measurement Study of Spino-Pelvic Parameters in Adolescent Idiopathic Scoliosis Patients","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2025-07-07 08:23:43","doi":"10.21203/rs.3.rs-6864546/v1","editorialEvents":[{"type":"communityComments","content":0},{"type":"editorInvitedReview","content":"","date":"2025-07-30T16:11:03+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"91177859817872756979179734718745857966","date":"2025-07-20T15:07:42+00:00","index":"hide","fulltext":""},{"type":"reviewersInvited","content":"","date":"2025-07-02T12:45:53+00:00","index":"","fulltext":""},{"type":"editorInvited","content":"","date":"2025-06-23T10:20:19+00:00","index":"","fulltext":""},{"type":"editorAssigned","content":"","date":"2025-06-12T05:36:51+00:00","index":"","fulltext":""},{"type":"checksComplete","content":"","date":"2025-06-12T05:36:33+00:00","index":"","fulltext":""},{"type":"submitted","content":"BMC Medical Imaging","date":"2025-06-10T15:16:25+00:00","index":"","fulltext":""}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"bmc-medical-imaging","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"bmim","sideBox":"Learn more about [BMC Medical Imaging](http://bmcmedimaging.biomedcentral.com/)","snPcode":"","submissionUrl":"https://www.editorialmanager.com/bmim/default.aspx","title":"BMC Medical Imaging","twitterHandle":"BMC_series","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"em","reportingPortfolio":"BMC Series","inReviewEnabled":true,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"f6f68c58-e754-4f5d-a4a1-34ed6389229c","owner":[],"postedDate":"July 7th, 2025","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"under-review","subjectAreas":[],"tags":[],"updatedAt":"2025-07-07T08:23:44+00:00","versionOfRecord":[],"versionCreatedAt":"2025-07-07 08:23:43","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-6864546","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-6864546","identity":"rs-6864546","version":["v1"]},"buildId":"8U1c8b4HqxoKbykW_rLl7","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}